WO1998058328A1 - Procede et dispositif de conversion de signal numerique, dispositif et procede de generation de matrice de transformation et support d'alimentation - Google Patents

Procede et dispositif de conversion de signal numerique, dispositif et procede de generation de matrice de transformation et support d'alimentation Download PDF

Info

Publication number
WO1998058328A1
WO1998058328A1 PCT/JP1998/002653 JP9802653W WO9858328A1 WO 1998058328 A1 WO1998058328 A1 WO 1998058328A1 JP 9802653 W JP9802653 W JP 9802653W WO 9858328 A1 WO9858328 A1 WO 9858328A1
Authority
WO
WIPO (PCT)
Prior art keywords
matrix
signal
digital signal
conversion
processing
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/JP1998/002653
Other languages
English (en)
French (fr)
Inventor
Hidetoshi Ichioka
Hideki Asazu
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sony Corp
Original Assignee
Sony Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sony Corp filed Critical Sony Corp
Priority to EP19980924648 priority Critical patent/EP0932106A1/en
Priority to CA 2263454 priority patent/CA2263454A1/en
Publication of WO1998058328A1 publication Critical patent/WO1998058328A1/ja
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • G06F17/14Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/102Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
    • H04N19/124Quantisation
    • H04N19/126Details of normalisation or weighting functions, e.g. normalisation matrices or variable uniform quantisers
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/48Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using compressed domain processing techniques other than decoding, e.g. modification of transform coefficients, variable length coding [VLC] data or run-length data
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/59Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving spatial sub-sampling or interpolation, e.g. alteration of picture size or resolution
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/60Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
    • H04N19/61Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding in combination with predictive coding
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/90Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using coding techniques not provided for in groups H04N19/10-H04N19/85, e.g. fractals
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/102Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
    • H04N19/124Quantisation

Definitions

  • the present invention relates to a frequency-domain digital signal subjected to orthogonal transform such as DCT (discrete cosine transform).
  • the present invention relates to a digital signal conversion method and apparatus for performing resolution conversion and base conversion, a conversion matrix creation method and apparatus, and a providing medium.
  • orthogonal transform coding such as DCT (Discrete Cosine Transform) coding has been known as a coding method for efficiently compressing still image data or moving image data. I have. When dealing with such orthogonally transformed digital signal, it may be necessary to change the resolution and the transformation basis.
  • the first orthogonally transformed digital signal with a resolution of 720 ⁇ 480 pixels as an example of a digital video format for home use, a so-called MPEG1 format of 35 2 ⁇ 240
  • the first signal is subjected to inverse orthogonal transformation to restore it to a time axis signal, and then interpolation is performed.
  • Orthogonal transformation is performed again by performing transformation processing such as decimation and thinning. In this way, the orthogonally transformed digital signal is often inversely transformed to restore the original signal, and then various transformation operations are performed, and then the orthogonal transformation is performed again.
  • orthogonal transformation and its inverse transformation require a lot of calculations, and when performing resolution conversion as described above, it is necessary to perform orthogonal transformation and inverse orthogonal transformation in addition to the resolution conversion processing. Can not perform good processing. In addition, there is a disadvantage that the signal is degraded because errors accumulate as the amount of calculation increases.
  • the present invention has been made in view of such circumstances, and has a digital signal conversion method that is simple in processing, causes little signal deterioration, and can perform conversion processing such as resolution conversion. It is an object of the present invention to provide a conversion matrix creating method and apparatus, and a providing medium.
  • the present invention generates an inverse orthogonal transform matrix T s ⁇ 1 for inverse orthogonal transform and an orthogonal transform matrix T d for orthogonal transform, and generates the generated inverse orthogonal transform.
  • a transformation matrix D is generated using the matrix T s ⁇ 1 and the orthogonal transformation matrix T d, and the input digital signal in the orthogonally transformed area is transformed in the area that is orthogonally transformed using the above transformation matrix D. And obtains the output digital signal of the orthogonally transformed area.
  • the input digital signal is obtained by orthogonally transforming the original signal in the time domain by the orthogonal transform matrix T s corresponding to the inverse orthogonal transform matrix T s ⁇ 1.
  • the output digital signal is decoded by an inverse transform matrix T d ⁇ 1 corresponding to the orthogonal transform matrix T d and converted into a signal in the time or space domain. To be converted.
  • the transformation matrix D is obtained by arranging m inverse orthogonal transformation matrices T s (R, 1 on the diagonal axis to form an L-order square matrix A, and an orthogonal transformation matrix T d of base length L (N low frequency basis vectors are taken out to create an N-by-L matrix B,
  • the transformation matrix D is a matrix that transforms the transformation basis. If the orthogonal transformation matrices T s and T d are different, orthogonal transformation is performed.
  • an n-dimensional digit signal is used as the input digit signal, and a conversion process is performed on the orthogonally transformed area using the transform matrix D for each dimension, and an n-dimensional output digit signal is obtained. Can be obtained.
  • the conversion magnification differs for each dimension, the conversion for each dimension A matrix may be created and conversion processing may be performed at different magnifications for each dimension using these conversion matrices.
  • the conversion block of the orthogonally transformed input digit signal is used as it is to compensate for the predetermined value to convert it to a block with a length of m times. What is necessary is just to obtain the output digital signal of the area that has been subjected to the orthogonal transformation.
  • FIG. 1 is an explanatory diagram for explaining a digitized signal conversion method according to a first embodiment of the present invention.
  • FIG. 2 is a diagram illustrating an example of input / output signals according to the first embodiment of the present invention.
  • FIG. 3 is a diagram illustrating an example of signals before and after conversion in a case where the first embodiment of the present invention is applied to a two-dimensional digital signal.
  • FIG. 4 is an explanatory diagram for explaining an operation of multiplying the input matrix signal by a conversion matrix.
  • FIG. 5 is an explanatory diagram for explaining a process of obtaining a two-dimensional output digital signal by applying a transformation matrix D twice to a two-dimensional input digital signal.
  • FIG. 6 is a flowchart illustrating an example of a processing procedure for creating a resolution conversion matrix.
  • FIG. 7 is a flowchart for explaining an example of a processing procedure for performing resolution conversion in the frequency domain using a resolution conversion matrix.
  • FIG. 8 is an explanatory diagram for explaining an enlargement process in the frequency domain.
  • FIG. 9 is a diagram for explaining conversion between two formats having different sampling ratios of component signals of a luminance signal and a color difference signal.
  • FIG. 10 is a flowchart for explaining an example of a processing procedure for creating a resolution conversion matrix in the case of a different reduction ratio for each dimension.
  • FIG. 11 is a flowchart for explaining an example of a processing procedure for performing resolution conversion in the frequency domain using a resolution conversion matrix in the case of a different reduction ratio for each dimension.
  • FIG. 12 is an explanatory diagram for explaining a digital signal conversion method according to the second embodiment of the present invention.
  • FIG. 13 is a diagram illustrating an example of input / output signals according to the second embodiment of the present invention.
  • Figure 14 shows the connection matrix A, the selection matrix W, the processing matrix X, and the partitioning matrix B. It is a figure for explaining each function.
  • FIG. 15 is a diagram for explaining an example in which an input signal is divided into metablocks and only necessary parts are selected and processed.
  • FIG. 16 is a flowchart for explaining the procedure for generating the connection matrix A.
  • Figure 17 is a flowchart for explaining the initialization procedure of the parameter.
  • FIG. 18 is a flowchart for explaining the procedure for generating the selection matrix W.
  • FIG. 19 is a flowchart for explaining the procedure for generating the processing matrix X.
  • FIG. 20 is a flowchart for explaining the procedure for generating the partitioning matrix B.
  • FIG. 21 is a diagram for explaining orthogonal transformation of a color difference signal of the DV (Digital Video) standard.
  • FIG. 22 is a diagram for explaining a resolution conversion process for a color difference signal of the DV (Digital Video) standard.
  • FIG. 23 is a block diagram showing a schematic configuration of hardware used in the embodiment of the present invention.
  • FIG. 24 is a system diagram showing an example of a system for converting the embodiment of the present invention from the DV standard to the MPEG1 standard.
  • FIG. 25 is a block diagram illustrating an example of a signal conversion device that converts the embodiment of the present invention from the DV standard to the MPEG-1 standard.
  • BEST MODE FOR CARRYING OUT THE INVENTION A digital signal conversion method and apparatus according to the present invention performs a conversion process such as a resolution conversion or a base conversion on an orthogonally transformed digital signal without performing an inverse orthogonal transformation or an orthogonal transformation again. Things.
  • the transformation matrix creation method according to the present invention creates a transformation matrix for directly performing the resolution conversion, the S-bottom transformation, and the like in an orthogonally transformed area, for example, a frequency domain or a sweat area. is there.
  • MPEG MPEG
  • ISO / IECJTC 1 / S C29 International Organization for Standardization / International Electrotechnical Commission, Joint Technical Committee 1 / Sub Committee 29. It is an abbreviation of Moving Picture Experts Group, which considers moving picture compression and coding in 29), and MP0EG172 is the MPEG1 standard and IS013818 is the MPEG2 standard.
  • IS011172-1 and IS013818-1 are for multimedia multiplexing
  • IS011172-2 and IS013818-2 are for video
  • IS011172-3 and IS013818-3 are for audio. Standardized.
  • the image signal is compressed and coded by using the correlation in the time and space direction of the image in the unit of a picture (frame or field).
  • the use of spatial correlation is realized by using DCT (Discrete Cosine Transform) coding.
  • orthogonal transform such as DCT is also performed by JPEG (Joint Photogra (Pic Coding Expert Group) is widely used for various image information compression coding.
  • JPEG Joint Photogra (Pic Coding Expert Group) is widely used for various image information compression coding.
  • orthogonal transform enables compression coding with high compression efficiency and excellent reproducibility by transforming an original signal in the time domain or the space domain into an orthogonally transformed domain such as the frequency domain ( hereinafter, referred to as the orthogonal transform).
  • Preferred embodiments of a digital signal conversion method and apparatus, a conversion matrix creation method and apparatus, and a providing medium according to the present invention will be described with reference to the drawings.
  • FIG. 1 is an explanatory diagram for explaining a digital signal conversion method and a conversion matrix creation method according to the first embodiment of the present invention.
  • a human-powered orthogonal transformation matrix generation unit 11 generates an inverse matrix T s ( k ) 1 of an orthogonal transformation matrix T s representing an orthogonal transformation performed in advance on an input digital signal, and generates a transformation matrix. It is sent to the generator 13.
  • the output orthogonal transformation matrix generator 12 generates an inverse transformation matrix T d (orthogonal transformation matrix T d corresponding to 1 ) indicating the inverse orthogonal transformation to be performed on the output digital signal.
  • Sent to 3 The conversion matrix generation unit 13 generates a conversion matrix D for performing conversion processing such as resolution conversion in the frequency domain, and sends it to the signal conversion unit 14.
  • the signal conversion unit 14 converts the input digital signal 15 converted into, for example, the frequency domain by the orthogonal conversion in an orthogonally converted area, for example, the frequency domain, and outputs the output digital signal.
  • Signal 16 shall be used.
  • an original signal in the time domain (or the spatial domain) (original signal A) is transformed into, for example, the frequency domain by the orthogonal transform matrix T s ( k ), and the frequency signal B i (corresponding to the above-mentioned input digital signal 15), which is reduced (or expanded) to, for example, N / L by the above-mentioned signal conversion section 14 to obtain a frequency signal B 2 (the above-mentioned output digital signal). It corresponds to No. 16.
  • the frequency signal B 2 is inversely orthogonally transformed by the inverse transformation matrix Td (L 1) to obtain a time-domain signal C.
  • the matrix (orthogonal transform matrix) in which the orthogonal transform base vector of length n is arranged in each row (orthogonal transform matrix) is T ( n ), and its inverse transform matrix is ⁇ Write like 1 .
  • T ( n ) the matrix (orthogonal transform matrix) in which the orthogonal transform base vector of length n is arranged in each row (orthogonal transform matrix)
  • ⁇ Write like 1 Denotes a vector representation of X.
  • both matrices are ⁇ -order square matrices.
  • the following equation (1) shows the one-dimensional DC ⁇ conversion matrix, 8) for ⁇ 28.
  • the inverse orthogonal transform matrix Ts ( k ) 1 generated by the input orthogonal transform matrix generator 11 corresponds to the orthogonal transform process (reverse process) when the input digital signal 15 is generated.
  • the orthogonal transformation matrix Td generated by the output orthogonal transformation matrix generation unit 12 (or the inverse of decoding the output digital signal converted by the signal conversion unit 14, that is, converting it to the time domain). It is assumed that these orthogonal transformation matrix generators 11 and 12 can generate base vectors of any length, corresponding to (or inverse of) the orthogonal transformation processing.
  • these orthogonal transformation matrix generators 11 and 12 may be the same orthogonal transformation matrix generator, in which case the orthogonal transformation matrices Ts ( k ) and Td ( L ) are It becomes the same kind of orthogonal transformation matrix that differs only in length.
  • the orthogonal transformation matrix generator exists for each different orthogonal transformation scheme.
  • the transformation matrix generation unit 13 converts the inverse orthogonal transformation matrix Ts (k) -1 generated by the input orthogonal transformation matrix generation unit 11 into a diagonal as shown in the following equation (2). Arrange m above, and create an L-order square matrix A.
  • the base length of the output digital signal 16 is N
  • the low-frequency base vectors N of the orthogonal transformation matrix Td are extracted, and a matrix B composed of N rows and L columns is created.
  • are N low-frequency components when Td (is represented by the basis vector as follows. ⁇
  • This matrix D is a conversion matrix for converting the resolution into the above-mentioned reduction ratio (or enlargement ratio) N / L.
  • Hi is a scalar value or a vector value, and is a coefficient for level correction and the like.
  • each metablock Mi is multiplied by a conversion matrix D.
  • the vector C i of length N obtained by this becomes a frequency vector signal in which each metablock is reduced to N / L.
  • a signal B 2 obtained by reducing the whole to N / L is obtained (the obtained output digital signal B 2 is an inverse orthogonal transform matrix Td ( N ) 1 can be converted to a time axis signal (C in Fig. 2).
  • the conversion matrix D shown in the above equation (5) is a conversion matrix for performing resolution conversion and base length conversion at a magnification of N / L.
  • N ⁇ L A matrix that transforms the transformation basis.
  • the output digit signal in the frequency domain uses the same decoder as the input digit signal, In other words, the same inverse orthogonal transform matrix can be used.
  • the transform matrix D does not transform the basis. Becomes a resolution conversion matrix that reduces the resolution to 1 / m.
  • the above operations are separable to the dimensions of the original signal, can be easily extended to more than two dimensions, and can be applied to various media such as audio, still images, and moving images .
  • the above-described conversion is first performed for each row, and then the same is applied to each column.
  • FIG. 3 schematically illustrates an example of each signal before and after conversion in a time domain and a frequency domain when the first embodiment of the present invention is applied to a two-dimensional digital signal such as an image signal.
  • FIG. The example of FIG. 3 shows a case where an original signal 21 which is a two-dimensional digital signal such as an image signal is reduced to 1/2 in each of the vertical and horizontal directions and converted into a two-dimensional digital signal 22. .
  • the original signal 21 is divided into blocks of, for example, 8 ⁇ 8 pixels in length and width, and an 8 ⁇ 8 matrix is used by an orthogonal transformation matrix T s ( 8 ) such as a DCT (discrete cosine transform) with a base length of 8.
  • T s orthogonal transformation matrix
  • the signal is converted into a signal 23 in the frequency domain in units of the blocks, and this is used as an input digital signal in the frequency domain.
  • This one block converts a base of length 8 into a base of length 16 for the two-dimensional input digital signal 23 composed of 8 ⁇ 8 frequency domain data by the base conversion matrix T as described above.
  • the digital signal 24 on the frequency axis with 16 x 16 blocks as one block, and the low-frequency portion of this 1 block 16 x 16 blocks indicated by oblique lines in the figure By extracting only 8 ⁇ 8 from each block and combining them, an output digital signal 25 in the frequency domain can be obtained.
  • the output digital signal 25 is converted into a time-domain digital signal 22 by an inverse orthogonal transformation matrix T d ( 8 ) 1 having a base length of 8. If it is decoded (decoded), this is a signal whose resolution has been converted so that the original signal 21 in the time domain is reduced to 1/4.
  • the transformation matrix D is an 8 ⁇ 16 matrix from the above equation (5). As shown in FIG. 4, this 8 ⁇ 16 conversion matrix D is composed of two blocks arranged vertically in the input domain signal 23 in the frequency domain. By multiplying by 2, the shaded portion (8 x 1) of the signal 26 whose vertical length is compressed to 1/2 is obtained. By performing this for all of the input digital signal 23, that is, by multiplying the matrix D for every two blocks of each column of the input digital signal 23, the vertical length becomes 1 / With 2, a signal 26 with the same horizontal length as the original is obtained.
  • a conversion matrix D is applied to the input digital signal 23 to obtain a signal (matrix data) 26 having a vertical length of 1/2.
  • the signal 26 is transposed, that is, the rows and columns are interchanged, and a transposed signal 26 1 is obtained.
  • the vertical length i.e. the original incoming The signal (corresponding to the horizontal length of the power signal signal 23) is reduced by half to obtain a signal (matrix data overnight) 27.
  • the obtained signal 27 1 becomes the above output digital signal 25 with both the length and width of the input digital signal 23 reduced to 1/2.
  • FIG. 6 is a flowchart showing a procedure of the above-described resolution conversion, in particular, a procedure of creating a conversion matrix D for reducing each dimension to 1 / m.
  • the length of the base used for the input signal is set to k, and the reduction ratio m (that is, reduced to 1 / m) is determined in the next step S32.
  • step S 35 As shown in the above equation (2), the m inverse transformation matrices T ( k) -1 are arranged on the diagonal to generate the L-order square matrix A.
  • step S 3 a matrix T of base length L, which is the output orthogonal transformation matrix described above, is created, and in step S37, k low-frequency base vectors of the matrix T are extracted.
  • Fig. 7 shows the resolution using the transformation matrix D created in the procedure of Fig. 6.
  • 9 is a flowchart showing a processing procedure when performing conversion.
  • the first step S41 in FIG. 7 it is determined whether or not the conversion has been completed in all dimensions of the input digital signal, and if the conversion has been performed in all dimensions, the resolution conversion is performed. The process ends.
  • the next step S42 the next dimension is set to the current dimension. Immediately after the start of resolution conversion, the first dimension is set to the current dimension.
  • step S43 it is determined whether or not the magnitude of the input signal is not equal to a multiple of L. If not, the process proceeds to step S44, where the input signal is supplemented with dummy data or the like to reduce L. To be a multiple of. In this way, the input signal that is originally a multiple of L or is a multiple of L by supplementation is divided into metablocks of length L in step S45.
  • the resolution can be reduced to 1 / m for all dimensions without changing the basis for a one-dimensional or multi-dimensional input digital signal.
  • the conversion matrix D can be used repeatedly during the resolution conversion processing of FIG. 7 once it is obtained by the procedure of FIG.
  • the resolution can be increased at an arbitrary magnification by adding a high-frequency component to the input digital signal in the frequency domain.
  • FIG. 8 shows an example of an enlargement method for supplementing 0 as a high-frequency component.
  • the original signal A in the time domain is subjected to orthogonal transformation such as DCT as described above to obtain an input digital signal B, in the frequency domain.
  • the input digital signal is expanded by m times for each block! ⁇
  • the original block is used as it is and zeros are filled in Signal B 2.
  • N / L reduction conversion may be performed using the digital signal after the enlargement as the input digital signal of the embodiment described above.
  • supplementing 0 to the high frequency component is an example, and a more appropriate value may be supplemented.
  • a plurality of conversion matrices generated at different sampling ratios with respect to the original signal such as a luminance signal and a color difference signal of a two-dimensional digital signal, are generated.
  • Embodiments of the present invention can also be applied to component signals.
  • FIG. 9 shows an example of conversion between two formats having different sampling ratios of components such as a luminance signal and a color difference signal.
  • the 8 ⁇ 8 blocks of the luminance signal 51 of the signal before conversion are grouped into four blocks, and one block (8 ⁇ 8) of each of the color difference signals 52 and 5 3 is provided. Yes, it is.
  • the luminance signal 56 block (8 x 8) of the converted signal has a total of 4 blocks each of 2 blocks vertically and horizontally.
  • One block (8 x 8) of each of the color difference signals 57 and 58 corresponds to the collected data.
  • the format is converted and the resolution is reduced to ⁇ each in the vertical and horizontal directions.
  • two types of resolution conversion matrices of 1/2 and 1/4 without changing the basis are prepared by the above-described resolution conversion process B of the present embodiment, and the input luminance signal 51 is both vertically and horizontally.
  • the output luminance signal 56 should be reduced to 1/2 and converted to the output luminance signal 56, and the input color difference signals 52 and 53 should be reduced to 1/4 only in the vertical direction.
  • the first step S61 of the flowchart of FIG. 10 it is determined whether or not the creation of the transformation matrix has been completed in all dimensions, and if the creation of the transformation matrix has been completed in all dimensions, the resolution conversion is performed. End the matrix generation process.
  • the length of the basis used for the input signal of the current dimension is set to k, and in the next step S64, the contraction rate ⁇ .,. / mi).
  • the order L.i of the next square matrix A, L, described later, is determined by Lj2kxmi.
  • the stearyl-up S 6 7 inverse transformation matrix is T, (to create a k.
  • the inverse transform matrix T (k) 1 Are arranged on a diagonal of mj to create a square matrix A of order L above.
  • the matrix T ( ) ) is extracted to create a matrix B consisting of k rows and L j columns.
  • a transformation matrix D j (two times jB A) is obtained from the above equation (5), where “hi” is a scalar single value or a vector value for correction in the j-th dimension.
  • FIG. 11 is a flowchart showing a processing procedure in the case where resolution conversion is performed for each dimension using the conversion matrices ⁇ , ⁇ created in the procedure of FIG.
  • the new dimension (j) is set to the current dimension.
  • step S83 it is determined whether or not the magnitude of the input signal is not equal to a multiple of Lj. To make it a multiple of L j. Thus, Lj is originally a multiple of L, or The multiplied input signal is divided into metablocks of length L j in step S85.
  • step S86 it is determined whether or not all the data blocks for the current dimension have been converted. If the conversion of all the data blocks has been completed, the flow returns to step S81, and all the meta data blocks have been processed. If the conversion of the block has not been completed, the process proceeds to step S87.
  • the resolution of the j-th dimension can be reduced to 1 / m j for the n-dimensional input digital signal without changing the basis.
  • the conversion matrix D. once obtained by the procedure shown in FIG. 10, can be used repeatedly during the resolution conversion process shown in FIG.
  • the transform basis can be arbitrarily changed for the input digitized signal transformed into the frequency domain by the orthogonal transform such as DCT,
  • the orthogonal transform such as DCT
  • the resolution and the like can be reduced at an arbitrary magnification.
  • the orthogonal transformation method and the overall length of the output digital signal being the same as the input digital signal, it is possible to obtain an output digital signal having a different base length. Can be.
  • the orthogonal transformation method, base length, and resolution Different output digital signals can be obtained.
  • the decoder for the input digital signal can be used as it is. Can be used and only resolution can be converted. Also, by combining enlargement and reduction, resolution conversion can be performed at any magnification. It can be applied to not only one-dimensional but also multi-dimensional frequency domain signals. In addition, since conversion can be performed at any magnification, correspondence between components is maintained even when the size or configuration of the conversion block differs depending on the component, such as a luminance signal or color difference signal in a two-dimensional signal. The resolution conversion can be performed as it is.
  • the compressed digital signal is subjected to only the compression / expansion, thereby enabling various transforms according to the first embodiment.
  • a metablock in which several blocks are put together is used, and resolution conversion can be performed without changing the block size.
  • the processing is performed on a part of the meta-block or the part that crosses the boundary of the meta-block. It cannot be performed, but by extending it as follows, resolution conversion can be performed at any magnification, and partial signals of any magnitude across block boundaries can be processed.
  • FIG. 12 is a diagram showing a digital device according to the second embodiment of the present invention.
  • FIG. 4 is an explanatory diagram for describing a signal conversion method and a conversion matrix creation method.
  • a concatenated matrix generation unit 1 1 1 that is an extension of the input orthogonal transformation matrix generation unit 11 of FIG. Section 112, a selection matrix generation section 117 for generating a selection matrix for extracting a part of the metablock or an arbitrary part crossing the boundary of the data block, and a desired resolution conversion processing.
  • a processing matrix generation unit 1 18 for generating a processing matrix for performing the signal processing of the above, and a conversion matrix generation unit 1 13 for obtaining a required matrix from these four matrices.
  • the transformation matrix D from 3 is sent to the signal transformation unit 114, and the orthogonally transformed input digital signal 115 is transformed in the orthogonal transformed state, and the output digital signal is output.
  • Each function of the division matrix B from the generation unit 112 will be described with reference to FIG.
  • (a) corresponds to the above-mentioned one metablock of the above-mentioned input digital signal 115, and this input digital signal (a) is obtained by using an orthogonally transformed domain, for example, a frequency domain or DCT. Compression This is the signal of the compression area that has been encoded.
  • the selection matrix W has a function of converting a part to be subjected to arbitrary processing from the signal (b) in the time or space domain.
  • the signal (c) with the length M from the p-th to the q-th of the signal L (b) in the time or space domain with the length L is extracted. This is for performing the actual signal conversion processing such as the above-described resolution conversion and reduction / enlargement in the time or space domain, and the user may specify a desired one.
  • the signal (c) of length M is converted to the signal (d) of length N.
  • the signal (e) of the area is used, and this signal (e) corresponds to the above-mentioned output digital signal 1 16.
  • the concatenated matrix generation unit 111 generates an inverse matrix Ts ( kI) -1 of an orthogonal transform matrix Ts ( kI ) 'representing an orthogonal transform previously performed on the input digital signal 1 15
  • the inverse matrix Ts ( kI ) 1 is arranged diagonally as shown in the following equation (7), and an L-order square matrix, A matrix of L rows and L columns (L kz Xim) is created, and this is referred to as a connection matrix A.
  • This equation (7) corresponds to the equation (2) in the first embodiment.
  • the selection matrix generation unit 117 generates a selection matrix W for selecting a part to be processed from the metablock as described above.
  • the selection matrix W is an element that is 1 for each column. There is one by one, and the rest is a matrix of 0.
  • the following equation (9) shows an M-by-L matrix W for extracting a signal of length M from the eye to the qth.
  • an M-order square matrix (a matrix of M rows and M columns) in which the M columns from the p columns to the q columns in the L columns are arranged diagonally with 1 is obtained. The remaining elements are all 0.
  • the processing matrix generation unit 118 generates a processing matrix X for actually performing signal processing, for example, resolution conversion.
  • the processing matrix X there is no need to be aware of the processing in the orthogonally transformed area, and the user may arbitrarily define a matrix that performs processing at the original signal level in the time or space domain.
  • the processing matrix X is a matrix of N rows and M columns in the example of FIG. 14 described above, and a specific example will be described later.
  • the transformation matrix generation unit 113 shown in FIG. 12 described above uses the following equation (based on the matrices generated by the matrix generation units 111, 117, 118, and 112). 1 0), an N-by-L transformation matrix D, i.e.,
  • the signal C i is a metablock having the length N described above, and has a length k. M of blocks. It is divided into pieces.
  • the processing matrix X may be a matrix for processing the original signal in the time or empty f?; J region without being aware that the signal to be processed is orthogonally transformed.
  • Various signal processing can be used. For example, resolution conversion processing of any magnification, processing of transforming the basis of orthogonal transformation, effect processing on compressed images such as JPEG and MPEG, filtering processing, digital video signal to MPEG And the like, and a format conversion process such as.
  • the processing matrix X any matrix can be used as long as resolution conversion can be performed on the original signal.
  • a matrix represented by the following equation (1 2) is used. . ⁇ in this equation (12) is a coefficient for level correction.
  • N LN Using these matrices, generate a transformation matrix D as described above, and perform signal conversion processing, so that the original signal is not restored, and the orthogonally transformed area remains unchanged at any magnification.
  • the resolution conversion can be performed.
  • the procedure of matrix generation in each of the matrix generation units 11 1, 1 17, 1 18, and 1 12 in FIG. 12 above This will be described with reference to the flow charts in FIGS. 16 to 20.
  • FIG. 16 is a flowchart for explaining a procedure of generating the connection matrix A in the connection matrix generation unit 111.
  • a first step S 12 parameters for matrix generation are initialized. The initialization of this parameter will be described later with reference to FIG.
  • the inverse orthogonal transform matrix Ts (k) 1 is created.
  • step S123 an arbitrary matrix of A and L rows and L columns is generated.
  • step S124 all elements of the matrix A are initialized to 0.
  • step S125 0 is set to 0. element and the matrices a 'diagonal the inverse orthogonal transform matrix on Ts (k) - 1 a and mi pieces arranged to produce a consolidated matrix a described above.
  • the entrance Block length k! And output block length k Doo is if different, the inverse orthogonal transform matrix that is created by the Sutetsu flop S 1 2 2 becomes T s 1 is a matter of course.
  • FIG. 17 is a flowchart showing an example of the procedure for initializing the parameters in step S121 of FIG.
  • the block length of input and output is k
  • the numerator a and denominator b of the above-mentioned resolution conversion magnification are set in steps S162 and S163, respectively.
  • L bxk is set in step S164
  • N aXk is set in step S165.
  • This initialization of parameters is also used in the procedure of generating each matrix shown in FIGS. 18 to 20 described below.
  • FIG. 18 is a flowchart for explaining the procedure of generating the selection matrix W in the selection matrix generation unit 117.
  • the initialization of the parameters described in FIG. 17 is performed in the first step S 13 1, and the unit matrix I of L rows and L columns is generated in the next step S 13 2.
  • the selection matrix W is set as this unit matrix I.
  • FIG. 19 is a flow chart for explaining the procedure for generating the processing matrix X in the processing matrix generation unit 118.
  • the parameters described in FIG. 17 are initialized, and in step S 142, an N ⁇ N inverse cosine transform matrix IDCT is obtained.
  • an L-by-L cosine transformation matrix DCT is obtained.
  • FIG. 20 is a flowchart for explaining a procedure of generating the partition matrix B in the partition matrix generation unit 112.
  • initialization of the parameters described in FIG. 17 is performed in the first step S 15 1, and the inverse orthogonal transformation matrix T du 1 is created in step S 15 2.
  • step S 1 5 3 generates an arbitrary matrix B ′ of N rows and N columns, initializes all elements of this matrix B ′ to 0 in step S 1 54, and in step S 1 5 5,
  • the above-described inverse orthogonal transformation matrix T d ( k 1 is arranged mo on the diagonal, and a division matrix B is generated.
  • the second embodiment can be easily extended to multidimensional signal processing.
  • the processing may be performed first in the vertical direction, and then performed in the horizontal direction.
  • any of the above-described resolution conversion processes can be extended to multidimensional signal processing, and can be used for resolution conversion of compressed image information such as so-called JPEG and MPEG.
  • format conversion will be described.
  • data may be stored in a partially irregular format for various reasons. For example, in the so-called DV (Digital Video) standard format, as shown in Fig. 21, a 4 x 8 pixel odd block at the right end of the chrominance signal can be created. It makes 8 8 blocks and decrypts it.
  • DV Digital Video
  • Such an irregular format can be processed without any problem by appropriately selecting the selection matrix W.
  • FIG. 23 is a block diagram showing an example of a schematic configuration of a hardware for realizing the above-described processing, particularly when a digital image signal is handled.
  • the CPU (Central Processing Unit) 101 for data calculation and signal processing is connected to the bus connected to the ROM 102, which stores programs and data, and is readable and writable.
  • RAM, image interface 104, image output interface 106, recording medium 108 such as a hard disk, etc. are connected, and digital data is input from terminal 105 via image input / output interface 104.
  • An image signal is input, and a digital image signal is output from the image output terminal 106 via the terminal 107.
  • the program used for the processing according to the embodiment of the present invention is recorded on the recording medium 108 such as the ROM 102 or a hard disk and provided. It can also be provided via a disk, such as a CD-ROM, or via a communication line.
  • video information captured by a home DV standard digital video camera 201 is converted to a digital video signal via an IEEE 1394 interface.
  • the signal of the DV standard has an enormous amount of data and is not suitable for handling in the personal computer 202. Therefore, using the above-described embodiment of the present invention, this DV standard video data is converted into MPEG 1 standard video data. By doing so, the amount of data can be significantly reduced, and non-linear editing by software within the personal computer 202 can be performed, and images can be transferred to other devices through the so-called in-network 203. It is possible to provide the information to 204.
  • the system conversion from the DV standard to the MPEG1 standard within the personal computer 202 is realized by the conversion processing unit 210 in FIG.
  • the supplied DV video data is variable-length decoded by the variable-length decoding unit 211 in the conversion processing unit 210, and is inversely quantized by the inverse quantization unit 211. Then, it is sent to the resolution converter 213 as it is in the orthogonally transformed area or the compressed area, and is subjected to resolution conversion.
  • the conversion process in the resolution conversion unit 21 is realized by the above-described embodiment. Also, after the motion prediction processing is performed on the area or the compressed area in the motion prediction section 214 which has been subjected to the orthogonal transformation, the motion is sent to the quantization section 215 and quantized.
  • the data is subjected to variable-length encoding in 216 and extracted as MPEG-1 video data.
  • the quantization unit 211 After the signal inversely quantized by the inverse quantization unit 211 is inversely orthogonally transformed into a time or space domain signal by the inverse DCT unit 211, The data format conversion is performed by the data conversion unit 222, the resolution conversion is performed by the resolution conversion unit 223, and the motion prediction is performed by the motion prediction unit 224 sequentially. Next, orthogonal transform is performed by the DCT unit 2 25 After transforming into an orthogonally transformed area or compressed area, the quantization unit
  • the processing matrix can be determined without being affected by the signal I: intersection basis, and the design of the matrix is easy and can be applied to various signal processing.
  • the inverse orthogonal transform matrix T s 1 for the inverse orthogonal transform and the orthogonal transform matrix T d for the orthogonal transform are generated, and the generated inverse orthogonal transform is generated.
  • a transformation matrix D is generated using the transformation matrix T s ⁇ 1 and the orthogonal transformation matrix T d, and the input digitized signal in the orthogonally transformed area is subjected to the orthogonal transformation using the above transformation matrix D in the domain.
  • the resolution and the transformation basis can be directly transformed in the orthogonally transformed area, and the decoding into the time domain and the spatial domain ( Inverse orthogonal transform) is not required, which simplifies calculations and enables high-quality conversion with small calculation errors.
  • the hardware burden is reduced, and the conversion process can be realized at a sufficiently high speed by software.
  • the input digital signal is a digital signal obtained by orthogonally transforming an original signal in a time domain or a space domain by an orthogonal transform matrix TS corresponding to the inverse orthogonal transform matrix T s ⁇ 1.
  • output di di Yuru signals, the orthogonal transform matrix T d to the corresponding inverse orthogonal transform matrix T cT 1 by the decoded time domain and it is elevation Gerare that shall be converted into a signal in the spatial domain also,
  • the original signal is orthogonally transformed by the orthogonal transformation matrix T s having a base length k for each transformation block of length k to obtain the input digital signal, and m input signals adjacent to the input digital signal are used.
  • the input digital signal can be reduced to N / L in the orthogonally transformed area.
  • N L
  • the transformation matrix D is a matrix that transforms the transformation basis.
  • orthogonal transformation matrices T s and T d are different, orthogonal transformation can be performed.
  • an n-dimensional digit signal is used as the input digit signal, and the transform is performed in the orthogonally transformed region using the transform matrix D for each dimension.
  • an n-dimensional output digital signal can be obtained.
  • the conversion magnification is different for each dimension, a conversion matrix for each dimension is created, and the conversion processing is performed at a different magnification for each dimension using these conversion matrices.
  • the conversion block of the input digital signal subjected to the orthogonal transformation is used as it is, and is converted to a block having a length of m times by filling a predetermined value.
  • At least one block of a fixed number of samples of the input digitized signal in the orthogonally transformed area is connected to form a data block, and a connection matrix for inverse orthogonal transformation is generated.
  • Generating a selection matrix for selecting and extracting a portion to be processed generating a processing matrix for performing a desired signal processing process on the selected portion, and orthogonalizing the signal processed signal.
  • a conversion matrix to be applied is generated, and the above-mentioned conversion matrix is applied to the orthogonally converted input digital signal for each of the above-mentioned blocks to obtain an output digital signal. Since the original signal does not need to be restored, it can be processed at high speed, and the signal is not degraded due to the dissolution error. Conversion can be performed, and the signal block size can be converted to an arbitrary size. Also, only a part of the signal can be selectively processed by the selection matrix, and in that case, it is not necessary to be aware of the boundaries of the blocks of the input signal. Also, the processing matrix can be determined without being affected by the orthogonal transform basis of the signal, and the matrix design is easy and can be applied to various signal processing.
  • the present invention is not limited to only the above-described embodiment.
  • resolution conversion of an image signal subjected to orthogonal transformation has been mainly described.
  • the present invention can be applied to digital signals in various frequency regions such as audio signals.
  • the orthogonally transformed signal is not limited to a signal in the frequency domain, and a signal in the time domain or the spatial domain is used as the original signal.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Algebra (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)
  • Complex Calculations (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)
  • Compression Of Band Width Or Redundancy In Fax (AREA)

Description

明 細 書 ディ ジ夕ル信号変換方法及び装置、 変換行列作成方法及び装置 並びに提供媒体 技 術 分 野 本発明は、 D C T (離散コサイン変換) 等の直交変換が施された 周波数領域のディ ジタル信号について、 解像度変換や基底変換を行 うためのディ ジ夕ル信号変換方法及び装置、 変換行列作成方法及び 装置並びに提供媒体に関するものである。 背 景 技 術 従来、 静止画データや動画デ一夕等を効率よく圧縮符号化するた めの符号化方式として、 D C T (離散コサイ ン変換) 符号化等の直 交変換符号化が知られている。 このような直交変換されたデイ ジ夕 ル信号を取り扱う際に、 解像度や変換基底を変更することが必要と されることがある。
例えば、 家庭用のディジタルビデオのフォーマツ 卜の一例として の 7 2 0 x 4 8 0画素の解像度の第 1の直交変換ディジ夕ル信号か ら、 いわゆる MP E G 1フォーマッ トの 3 5 2 x 2 40画素の解像 度の第 2の直交変換ディジ夕ル信号に変換する場合には、 第 1の信 号に対して逆直交変換を行って時間軸の信号に復元した後に、 補間 や間引き等の変換処理を行って、 再び直交変換を施している。 この ように、 直交変換されたディ ジタル信号は、 一旦逆変換されて原信 号を復元した後に各種変換操作が行われ、 その後再度直交変換され ることが多い。
ところで、 一般に直交変換及びその逆変換には多くの計算を要し、 上述したように解像度変換を行う場合には、 解像度変換の処理以外 に、 直交変換及び逆直交変換を行う必要があり、 効率のよい処理が 行えない。 また、 計算量の増加に伴い誤差が蓄積されるため、 信号 が劣化するという欠点もある。 発 明 の 開 示 本発明は、 このような実情に鑑みてなされたものであり、 処理が 簡単で、 信号劣化も少なく、 解像度変換等の変換処理が行い得るよ うなディ ジ夕ル信号変換方法及び装置、 変換行列作成方法及び装置 並びに提供媒体を提供することを目的とする。
本発明は、 上述した課題を解決するために、 逆直交変換のための 逆直交変換行列 T s—1 及び直交変換のための直交変換行列 T d を生 成し、 生成された上記逆直交変換行列 T s—1 及び直交変換行列 T d を用いて変換行列 Dを生成し、 直交変換された領域の入力ディジ夕 ル信号に対して上記変換行列 Dを用いて直交変換された領域で変換 処理を行い、 直交変換された領域の出力ディ ジ夕ル信号を得ること を特徴としている。
ここで、 上記入力ディ ジタル信号は、 時間領域の原信号を上記逆 直交変換行列 T s—1 に対応する直交変換行列 T s により直交変換し て得られた領域のディ ジ夕ル信号であり、 上記出力ディジ夕ル信号 は、 上記直交変換行列 T d に対応する逆 交変換行列 T d—1 により 復号されて時間あるいは空間領域の信号に変換されるものとするこ とが挙げられる。
また、 上記原信号を長さ kの変換プロ、ソク itf:に基底の長さ kの上 記直交変換行列 T s ( k )により直交変換して上記入力デイジ夕ル信号 とし、 この入力ディジ夕ル信号の隣接する m個のプロックから成る 長さ L ( = k X m ) の連続する信号を、 長さ Nのプロック 1個に変 換することが好ましい。
上記変換行列 Dは、 上記逆直交変換行列 T s ( R , 1を m個対角軸上 に配置して L次の正方行列 Aを作成し、 基底の長さ Lの直交変換行 列 T d (いの低周波基底べク トル N個を取り出して N行 L列の行列 B を作成し、
D = ひ · B · A
ただし、 ひはスカラー値又はべク トル値のレベル補正のための 係数
の式により作成することが挙げられる。
これにより、 直交変換された領域で、 入力ディジタル信号を N / Lに縮小変換することができる。 特に N = Lのとき、 変換行列 Dは、 変換基底を変換する行列になる。 また、 直交変換行列 T s と T d と が異なる場合には、 直交変換方式の変換を行うことになる。
また、 上記入力ディジ夕ル信号として n次元のディジ夕ル信号を 用い、 各次元毎に上記変換行列 Dを用いて直交変換された領域で変 換処理を行い、 n次元の出力ディジ夕ル信号を得ることができる。 このとき、 各次元毎に変換倍率が異なる場合には、 各次元毎の変換 行列を作成してこれらの変換行列を用いて各次元毎に異なる倍率で 変換処理を行わせればよい。
解像度の拡大処理を直交変換された領域で行う場合には、 直交変 換された入力ディジ夕ル信号の変換プロックをそのまま用いて所定 の値を補うことにより長さが m倍のプロックに変換し、 直交変換さ れた領域の出力ディジタル信号を得るようにすればよい。
また、 本発明によれば、 直交変換された領域の入力ディ ジタル信 号の一定サンプル数のプロックを少なく とも 1つ連結してメ夕プロ ックを形成すると共に逆直交変換する連結行列を生成し、 上記メ夕 プロックから処理の対象とすべき部分を選択して取り出す選択行列 を生成し、 上記選択された部分に対して所望の信号加工処理を施す 加工行列を生成し、 上記信号加工処理された信号に対して直交変換 を施すと共に少なく とも 1つのブロックに分割する機能を有する分 割行列を生成し、 上記連結行列、 選択行列、 加工行列、 分割行列を 用いて直交変換された領域の信号に対して変換処理を施すための変 換行列を生成し、 直交変換された入力ディジ夕ル信号に対して上記 メ夕プロック毎に上記変換行列を適用して出力ディ ジ夕ル信号を得 ることを特徴とする。 図面の簡単な説明 図 1は、 本発明の第 1の実施の形態となるディジ夕ル信号変換方 法を説明するための説明図である。
図 2は、 本発明に係る第 1の実施の形態における入出力信号の一 例を示す図である。 図 3は、 2次元のディジ夕ル信号に本発明の第 1の実施の形態を 適用する場合の変換前後の信号の一例を示す図である。
図 4は、 変換行列を入力ディ ジ夕ル信号に対して掛ける操作を説 明するための説明図である。
図 5は、 2次元の入力ディ ジ夕ル信号に対して変換行列 Dを 2回 作用させることにより 2次元の出力ディジ夕ル信号を得る処理を説 明するための説明図である。
図 6は、 解像度変換行列を作成する処理手順の一例を説明するた めのフローチャートである。
図 7は、 解像度変換行列により周波数領域で解像度変換を行う処 理手順の一例を説明するためのフロ一チヤ一トである。
図 8は、 周波数領域での拡大処理を説明するための説明図である c 図 9は、 輝度信号及び色差信号の各コンポーネン ト信号のサンプ リング比が異なる 2つのフォーマッ ト間の変換を説明するための説 明図である。
図 1 0は、 次元毎に異なる縮小率の場合の解像度変換行列を作成 する処理手順の一例を説明するためのフローチャー トである。
図 1 1は、 次元毎に異なる縮小率の場合の解像度変換行列により 周波数領域で解像度変換を行う処理手順の一例を説明するためのフ ローチャートである。
図 1 2は、 本発明の第 2の実施の形態となるディ ジ夕ル信号変換 方法を説明するための説明図である。
図 1 3は、 本発明に係る第 2の実施の形態における入出力信号の 一例を示す図である。
図 1 4は、 連結行列 A、 選択行列 W、 加工行列 X、 分割行列 Bの それぞれの機能を説明するための図である。
図 1 5は、 入力信号をメタプロックに分割して必要な部分のみを 選択して処理する場合の例を説明するための図である。
図 1 6は、 連結行列 Aの生成手順を説明するためのフロ一チヤ一 トである。
図 1 7は、 パラメ一夕の初期化手順を説明するためのフローチヤ —トである。
図 1 8は、 選択行列 Wの生成手順を説明するためのフローチヤ一 トである。
図 1 9は、 加工行列 Xの生成手順を説明するためのフ口一チヤ一 トである。
図 2 0は、 分割行列 Bの生成手順を説明するためのフローチヤ一 トである。
図 2 1は、 D V (ディジタルビデオ) 規格の色差信号の直交変換 を説明するための図である。
図 2 2は、 D V (ディジタルビデオ) 規格の色差信号に対する解 像度変換処理を説明するための図である。
図 2 3は、 本発明の実施の形態に用いられるハードウエアの概略 構成を示すブロック図である。
図 2 4は、 本発明の実施の形態を、 D V規格から M P E G 1規格 に変換するシステムの一例を示すシステム図である。
図 2 5は、 本発明の実施の形態を、 D V規格から M P E G 1規格 に変換する信号変換装置の一例を示すプロック図である。 発明を実施するための最良の形態 本発明に係るディ ジ夕ル信号変換方法及び装置は、 直交変換され たディ ジタル信号を、 逆直交変換や再度の直交変換を伴わずに、 解 像度変換や基底変換等の変換処理を施すものである。 また、 本発明 に係る変換行列作成方法は、 上記解像度変換や S底変換等を直交変 換された領域、 例えば周波数領域あるいは汗縮領域で直接的に行わ せるための変換行列を作成するものである。
直交変換符号化を含む画像符号化の具体例としては、 いわゆる M P E Gの符号化規格が挙げられる。 この MP E Gとは、 I S O/ I E C J T C 1 /S C29 ( International Organization for Stand ardization /International Electrotechnical Commission, Joint Technical Committee 1 /Sub Committee 29: 国際標準化機構/国 際電気標準会議 合同技術委員会 1 /専門部会 2 9 ) の動画像圧縮 符号化の検討組織 (Moving Picture Experts Group) の略称であり、 MP E G 1標準として IS0U172が、 MP E G 2標準として IS013818 がある。 これらの国際標準において、 マルチメディア多重化の項目 で IS011172- 1及び IS013818- 1が、 映像の項目で IS011172-2及び IS01 3818- 2が、 また音声の項目で IS011172- 3及び IS013818-3がそれぞれ 標準化されている。
ここで、 画像圧縮符号化規格としての IS011172- 2又は IS013818-2 においては、 画像信号を、 ビクチャ (フレーム又はフィールド) 単 位で、 画像の時間及び空間方向の相関を利用して、 圧縮符号化を行 つており、 空間方向の相関の利用は、 D C T (離散コサイ ン変換 : Discrete Cosine Transform) 符号化を用いることで実現している。
この D C T等の直交変換は、 この他、 J P E G (Joint Photogra phi c Coding Expert Group) 等の種々の画像情報圧縮符号化に広く 採用されている。 一般に直交変換は、 時間領域あるいは空間領域の 原信号を周波数領域等の直交変換された領域に変換することにより、 圧縮効率が高く再現性の優れた圧縮符号化を可能にするものである ( 以下、 本発明に係るディジタル信号変換方法及び装置、 変換行列 作成方法及び装置並びに提供媒体の好ましい実施の形態について図 面を参照しながら説明する。
図 1は、 本発明の第 1の実施の形態となるディ ジ夕ル信号変換方 法及び変換行列作成方法を説明するための説明図である。
この図 1において人力直交変換行列生成部 1 1では、 入力ディ ジ タル信号に対して予め施された直交変換を表す直交変換行列 T s の逆行列 T s (k ) 1を生成し、 変換行列生成部 1 3に送っている。 出 力直交変換行列生成部 1 2では、 出力ディジタル信号に対して施さ れる逆直交変換を示す逆変換行列 T d (い 1に対応する直交変換行列 T d )を生成し、 変換行列生成部 1 3に送っている。 変換行列生成 部 1 3では、 解像度変換等の変換処理を周波数領域で行うための変 換行列 Dを生成し、 信号変換部 1 4に送る。 この信号変換部 1 4は、 直交変換により例えば周波数領域に変換された入力ディ ジ夕ル信号 1 5を、 例えば周波数領域等の直交変換された領域のままで変換処 理して、 出力ディ ジタル信号 1 6 とするものである。
すなわち、 例えば図 2に示すように、 元の時間領域 (あるいは空 間領域) の信号 (原信号 A ) を、 上記直交変換行列 T s (k)により例 えば周波数領域に変換して周波数信号 B i (上記入力ディジ夕ル信 号 1 5に相当) とし、 これを上記信号変換部 1 4により例えば N / Lに縮小 (又は拡大) して周波数信号 B 2 (上記出力ディ ジタル信 号 1 6に相当する。 ) とし、 この周波数信号 B 2 を上記逆変換行列 Td(L 1により逆直交変換して、 時間領域の信号 Cを得るようにし ている。
ここで、 図 2の例では、 1次元の原信号 Aを、 長さ kの変換プロ ック毎に直交変換し、 得られた周波数領域の変換プロックの隣接す る m個のブロック、 すなわち長さ L ( = k X m) の連続する周波数 信号を、 長さ N (ただし、 N< L) の 1つのプロックに変換する場 合、 すなわち全体を N / Lに縮小する場合を示している。
以下の説明では、 長さ nの直交変換基底べク トルく旦,, 2 , ··· , n >を各行に配列した行列 (直交変換行列) を T (n) 、 その逆変換 行列を Τ 1 のように記述する。 なお、 は、 Xのベク トル表現 を示す。 このとき、 いずれの行列も η次の正方行列である。 一例と して、 η二 8のときの 1次元 D C Τ変換行列 Τ ,8) を、 次の式 ( 1 ) に示す。
ί- ( 1/V2 '
£2.
Figure imgf000012_0001
cos(15W16) f3 cos(2^/l6) cos(6^/ 16) cos(l0?r/l6) cos(14Jr/l6) cos(18 /16) cos(22^/16) cos(26^/16) cos(30^/16)
£4 1 cos(3^r/ 16) cos(9^-/16) cos (15-^/ 16) cos(21^/16) cos(27 /16) cos(33^/16) cos(39^/16) COS(45 /16)
((8)
fs 2 cos(4Jr/16) cos(l2^/i6) cos(2CPr/16) cos(28Jr/16) cos(36^/16) cos(44^/i6) cos(52^/ 16) cos(60^r/16)
£6 cos(5^/16) cos(15兀 / 16) cos(25^/16) COS(35JT/16) cos(45ir/16) cos(55^r/16) COS(65JT/16) cos(75Jr/16)
£7 cos(6^/16) cos(l8^r/16) cos(30 /16) cos(42^/16) cos(54^/16) cos(66^/i6) cos(78^/16) COS(90JT/16) cos(7兀 /16) cos(21^r/16) cos(35丌 /16) cos(49^/16) cos(63^r/16) co つ Ίπ 116) cos(9i^/l6) COS(105JT/16) £s a)
上記図 1において、 既に直交変換行列 Ts (k)により周波数領域に 直交変換された入力ディ ジ夕ル信号 1 5について、 その直交変換ブ ロックの大きさ、 すなわち基底の長さが kであるとき、 上記入力直 交変換行列生成部 1 1により逆直交変換行列 Ts(k) — 1を生成し、 ま た、 上記出力直交変換行列生成部 1 2により基底の長さが L ( = k X m ) の直交変換行列 Tdu.)を生成する。
このとき、 入力直交変換行列生成部 1 1により生成される逆直交 変換行列 Ts(k) 1は、 入力ディ ジ夕ル信号 1 5を生成する際の直交 変換処理 (の逆処理) に対応し、 出力直交変換行列生成部 1 2によ り生成される直交変換行列 Td (いは、 信号変換部 1 4で変換された 出力ディジタル信号を復号する際、 すなわち時問領域に変換する際 の逆直交変換処理 (の逆処理) に対応し、 これらの直交変換行列生 成部 1 1、 1 2共に、 任意の長さの基底べク トルを生成することが できるものとする。
なお、 これらの直交変換行列生成部 1 1、 1 2は、 同一の直交変 換行列生成部であってもよく、 この場合、 直交変換行列 Ts(k)と Td(L)とは、 基底の長さのみ異なる同一種の直交変換行列になる。 直交変換行列生成部は、 異なる直交変換方式毎に存在するものであ る。
次に、 変換行列生成部 1 3においては、 入力直交変換行列生成部 1 1により生成された上記逆直交変換行列 Ts(k) — 1を、 次の式 ( 2) に示すように、 対角上に m個配置して、 L次の正方行列 Aを作成す る。 また、 出力ディジタル信号 1 6の基底の長さを Nとするとき、 上記直交変換行列 Td ( の低周波基底べク トル N個を取り出し、 N 行 L列から成る行列 Bを作成する。
Figure imgf000014_0001
ただし、 …, ^は、 Td (いを以下のように基底ベク トル で表したとき、 低周波の N個を取り出したものである。 ヽ
-- (斗)
Figure imgf000014_0002
そして、
D =ひ · Β . Α · · · ( 5 )
を計算し、 Ν行 L列の行列 Dを作成する。 この行列 Dが、 上記縮小 率 (又は拡大率) N/Lに解像度を変換する変換行列になる。 なお、 ひはスカラー値又はべク トル値で、 レベル補正等のための係数であ る。
上記図 1の信号変換部 14において、 図 2に示すように、 周波数 領域の入力ディジ夕ル信号 B 1 のプロック m個をひとまとめにし、 Lの大きさのメタブロック ( 1メタブロック = mブロック) に分割 する。 入力ディジタル信号 の長さが Lの倍数でない場合には、 信号を補う等により、 例えば 0等のダミーデータを充填 (スタッフ イ ング) すること等により、 Lの倍数になるようにする。 このよう にしてできたメタブロックを M i ( i = 0 , 1 , 2, · · · ) とす る。
図 1の信号変換部 1 4においては、 上記各メタブロック Mi に変 換行列 Dを掛ける。
C i 二 D · Mi · · · ( 6 )
これにより得られた長さ Nのベク トル C i が、 各メタブロックを N /Lに縮小した周波数べク トル信号になる。 全メタブロックに対し てこれを行えば、 全体を N/Lに縮小変換した信号 B2 が得られる ( 得られた出力ディジタル信号 B 2 は、 基底の長さ Nの逆直交変換 行列 Td(N) 1によって、 時間軸の信号 (図 2の C) に変換すること ができる。
ここで、 N≠kの場合、 上記式 ( 5) に示す変換行列 Dは、 N/ Lの倍率で解像度変換及び基底の長さの変換を行う変換行列になり、 特に N二 Lのとき、 変換基底を変換する行列になる。 また、 上記直 交変換行列 Ts(k)と Td )とが異なる場合、 直交変換方式の変換に なる。
特に、 上記直交変換行列 Ts(k)と Td (い とが同一でかつ N二 kと する場合、 周波数領域での上記出力ディジ夕ル信号は入力ディジ夕 ル信号と同じ復号器を用いること、 すなわち同じ逆直交変換行列を 用いることができ、 この場合、 上記変換行列 Dは、 基底を変換せず に解像度を 1 / mに縮小する解像度変換行列になる。
以上の操作は、 原信号の次元に対して分離可能であり、 2次元以 上の場合にも容易に拡張することができ、 音声、 静止画、 動画等の 種々のメディアに適用することができる。 例えば、 2次元信号に対 しては、 先ず各行に対して上述の変換を施した後、 各列に対して同 様に適用すればよい。
図 3は、 画像信号のような 2次元のディジ夕ル信号に本発明の第 1の実施の形態を適用する場合の時間領域、 周波数領域における変 換前後の各信号の一例を概略的に示す図である。 この図 3の例では、 例えば画像信号等の 2次元ディジタル信号である原信号 2 1を、 縦 横それぞれ 1 / 2に縮小して 2次元ディジ夕ル信号 2 2に変換する 場合を示している。
この図 3において、 原信号 2 1を例えば縦横 8 x 8画素のプロッ クに分割し、 基底の長さ 8の D C T (離散コサイ ン変換) 等の直交 変換行列 T s ( 8 )により 8 X 8のブロックを単位とする周波数領域の 信号 2 3に変換して、 これを周波数領域の入力ディ ジ夕ル信号とす る。 この 1ブロックが 8 X 8個の周波数領域データから成る 2次元 入力ディジタル信号 2 3に対して、 上述したような基底変換行列 T により、 長さ 8の基底を長さ 1 6の基底に変換するような基底変換 を施して、 1 6 X 1 6個を 1 ブロックとする周波数軸のディジ夕ル 信号 2 4とし、 この 1 ブロック 1 6 x 1 6個の内の図中斜線で示す 低周波部分の 8 x 8個のみをそれぞれのブロックから取り出してま とめることにより、 周波数領域の出力ディ ジ夕ル信号 2 5が得られ る。 この出力ディ ジタル信号 2 5に対して、 基底の長さ 8の逆直交 変換行列 T d ( 8 ) 1によって時間領域のディジ夕ル信号 2 2に変換 (復号) すれば、 これが上記時間領域の原信号 2 1を 1 / 4に縮小 するように解像度変換した信号となる。
ここで、 図 3においては、 長さ 8の基底を長さ 1 6の基底に変換 した際に得られるディ ジ夕ル信号の内の低周波べク トル 8 X 8個を 取り出す計算を、 直接に、 変換行列 Dを作 fflさせるだけで得られる ことを示している。 ただし、 2次元の周波数信 の縦横に適用する ために、 Dを 2回掛けることが必要とされる。 この変換行列 Dの適 用例について、 図 4及び図 5を参照しながら説明する。
この具体例において、 周波数領域の入力ディ ジ夕ル信号の基底の 長さ kは 8で、 メタブロックの長さ Lは、 m = 2より、 L 二 8 x 2 = 1 6である。 また、 周波数領域の出力ディ ジタル信号の基底の長 さ Nは 8であるから、 上記式 ( 5 ) より、 変換行列 Dは 8 X 1 6の 行列となる。 この 8 X 1 6の変換行列 Dを、 図 4に示すように、 周 波数領域の入力ディ ジ夕ル信号 2 3の縦に 2ブロックまとめたもの の 1列分 ( 1 6 X 1 : 図中の斜線部) に掛けることにより、 縦が 1 / 2に圧縮された信号 2 6の斜線部 ( 8 x 1 ) が得られる。 これを 入力ディ ジ夕ル信号 2 3の全てに対して行うことにより、 すなわち 入力ディジ夕ル信号 2 3の各列の 2プロック毎に行列 Dを掛けるこ とにより、 縦の長さが 1 / 2で、 横の長さが元と同じ信号 2 6が得 られる。
すなわち図 5に示すように、 入力ディジ夕ル信号 2 3に変換行列 Dを作用させて、 縦の長さが 1 / 2の信号 (マ ト リクスデータ) 2 6を得る。 この信号 2 6を転置して、 すなわち行と列とを入れ替え て、 転置信号 2 6 1 を得る。 この転置信号 2 6 1 に対して、 再び上 記変換行列 Dを作用させることにより、 縦の長さ (すなわち元の入 力ディ ジ夕ル信号 2 3の横の長さに相当) を 1 /2にした信号 (マ ト リクスデ一夕) 2 7を得る。 この信号 2 7を転置して縦横を元に 戻すと、 得られた信号 2 71 は入力ディ ジ夕ル信号 2 3の縦横を共 に 1 / 2にした上記出力ディ ジ夕ル信号 2 5 となる。
以上説明した本発明の実施の形態の処理をまとめて、 一般に 1次 元又は複数次元の入力ディジ夕ル信号の全ての次元について解像度 を 1 /mに縮小する場合の例について、 図 6及び図 7のフ口一チヤ 一トを参照しながら説明する。
図 6は、 上述した解像度変換、 特に各次元毎にそれぞれ 1 /mに 縮小するための変換行列 Dの作成の手順を示すフローチヤ一トであ る。 この図 6のフローチャートの最初のステップ S 3 1では、 入力 信号に用いられている基底の長さを kとし、 次のステップ S 3 2で 縮小率 m (すなわち 1 /mに縮小) を決定している。 次のステップ S 3 3では、 後述する L次の正方行列 Aの次数 Lを、 L = k xmに より求める。 次のステップ S 3 4では、 長さ kの直交変換基底べク トルく !, 2,"', 1< >を各行に配列した行列 (直交変換行列) を Τ (, とするとき、 その逆変換行列である T )—1 を作成する。 次 のステップ S 3 5では、 上記式 ( 2 ) に示したように、 上記逆変換 行列 T (k)1 を m個対角上に配置して、 上記 L次の正方行列 Aを作 成する。 次に、 ステップ S 3 6で、 上述した出力直交変換行列とな る基底の長さ Lの行列 T ( を作成し、 次のステップ S 3 7で、 こ の行列 T ) の低周波基底ベク トル k個を取り出して、 k行 L列か ら成る行列 Bを作成する。 最後のステップ S 3 8で、 上記式 ( 5 ) — より、 変換行列 D ( = a A) を求める。
次に図 7は、 図 6の手順で作成された変換行列 Dを用いて解像度 変換を行う場合の処理手順を示すフローチヤ一トである。 この図 7 の最初のステツプ S 4 1では、 入力ディジ夕ル信号の次元数の全て の次元で変換が終了したか否かを判別しており、 全ての次元で変換 していれば、 解像度変換処理を終了する。 次のステップ S 4 2では、 次の次元を現在の次元に設定する。 なお、 解像度変換開始直後では、 第 1の次元が現在の次元に設定されることは勿論である。
次のステツプ S 4 3では、 入力信号の大きさが Lの倍数に等しく ないか否かを判別し、 等しくないときステップ S 4 4に進んで、 入 力信号にダミーデータ等を補足して Lの倍数になるようにしている。 このように、 元々 Lの倍数の、 あるいは補足により Lの倍数とされ た入力信号は、 ステップ S 4 5により、 長さ Lのメタブロックに分 割される。
次のステップ S 4 6では、 現在の次元についての全メ夕プロック を変換処理したか否かを判別し、 全メ夕プロックの変換が終了して いるときには上記ステップ S 4 1に戻り、 全メタプロックの変換が 終了していないときにはステップ S 4 7に進んでいる。 ステップ S 4 7では、 次のメ夕ブロヅクを現在のメ夕ブロックべク トル Mに設 定し、 ステップ S 4 8で現在のメタブロックに上記変換行列 Dを掛 ける、 すなわち C = D Mを計算し、 ステップ S 4 9でこの Cを出力 して、 上記ステップ S 4 6に戻っている。
以上の図 6、 図 7に示した手順により、 1次元又は複数次元の入 力ディ ジタル信号に対して、 基底を変えずに、 全ての次元について 解像度を 1 / mに縮小することができる。 なお、 変換行列 Dは、 図 6の手順により一度求めておけば、 図 7の解像度変換処理中は反復 して使用できることは勿論である。 以上の説明においては、 主として縮小の向きに解像度変換を行う 例について述べたが、 拡大も可能である。 すなわち、 一般に、 周波 数領域の入力ディジ夕ル信号に対して、 高周波成分を追加すること で、 任意の倍率で解像度を拡大することができる。
図 8に、 高周波成分として 0を補う拡大方法の一例を示している。 この図 8において、 時間領域の原信号 Aに上述したような D C T等 の直交変換を施して周波数領域の入力ディ ジ夕ル信号 B , を得る。 の入力ディジタル信号 を、 各ブロック毎に m倍に拡大する!^祭 に、 元のブロックをそのまま用い、 拡大されたブロックの残りの部 分には 0を埋めることにより、 周波数領域の出力ディジ夕ル信号 B 2 を得ている。 または、 拡大後のディジタル信号を、 上述した実 施の形態の入力ディ ジ夕ル信号として、 N / L縮小変換を行うよう にしてもよい。 なお、 高周波成分に 0を補うのは一例であって、 よ り適切な値を補ってもよい。
ところで、 複数の異なる倍率、 縮小率の変換行列を作成すること により、 例えば 2次元ディジ夕ル信号の輝度信号や色差信号等のよ うに、 原信号に対して異なるサンプリング比で生成された複数のコ ンポ一ネン ト信号に本発明の実施の形態を適用することも可能であ る。
図 9は、 輝度信号、 色差信号等の各コンポーネン トのサンプリン グ比が異なる 2つのフォーマツ ト間の変換の例を示している。 この 図 9において、 変換前の信号の輝度信号 5 1の 8 X 8のブロックが 横に 4ブロックまとめられたものに対して、 色差信号 5 2、 5 3の 各 1ブロック ( 8 x 8 ) が対応している。 また、 変換後の信号の輝 度信号 5 6のブロック ( 8 x 8 ) が縦横 2ブロックずつ計 4ブ口ヅ クまとめられたものに対して、 色差信号 5 7、 5 8の各 1ブロック ( 8 x 8 ) が対応している。
従って、 この図 9に示す変換処理においては、 フォーマッ トを変 換すると共に解像度を縦横それぞれ 1 /2に圧縮している。 この場 合、 上述した本実施の形態の解像度変換処 B により、 基底を変えな い 1 /2と 1 /4の 2種類の解像度変換行列を用意し、 入力輝度信 号 5 1については縦横共に 1 / 2に縮小して出力輝度信号 5 6に変 換し、 入力色差信号 5 2、 5 3については縦方向についてのみ 1 / 4に縮小変換すればよい。
次に、 次元毎に異なる縮小率、 すなわち一般的には次元毎に任意 の倍率、 縮小率の場合の解像度変換処理手順について、 図 1 0及び 図 1 1を参照しながら説明する。
図 1 0は、 上述した各次元毎にそれぞれ任意の縮小率で縮小する ための変換行列の作成の手順を示すフローチヤ一トである。 すなわ ち、 n次元の各次元毎に任意の縮小率 m., ( j =1,2,···, n) で縮小 するための変換行列 Dj の作成の手順を示している。 この場合、 各 次元の入力ディ ジタル信号のメタプロックの長さをそれぞれ L j ( j = 1,2,···,η) とし、 また変換行列を Dj ( j =1,2,···, n) としてい る。
この図 1 0のフローチャートの最初のステップ S 6 1では、 全て の次元で変換行列の作成が終了したか否かを判別し、 全ての次元で 変換行列の作成が終了していれば、 解像度変換行列の生成処理を終 了する。 次のステップ S 6 2では、 新しい次元を現在の次元 ( j ) に設定する。 なお、 解像度変換行列作成処理の開始直後では、 第 1 の次元 ( j = l ) が現在の次元に設定されることは勿論である。 次のステップ S 6 3では、 現在の次元の入力信号に用いられてい る基底の長さを kとし、 次のステップ S 6 4で、 現在の次元での縮 小率 πι.,· (すなわち 1 /m.iに縮小) を決定している。 次のステップ S 6 5では、 後述する L ,, 次の正方行列 Aの次数 L .i を、 L j二 k x m.iにより求める。 次のステップ S 6 6では、 上述した直交変換行列 T d に対しての逆変換行列である T (k を作成する。 次のステ ップ S 6 7では、 上記逆変換行列 T (k) 1 を mj個対角上に配置して、 上記 L」次の正方行列 Aを作成する。 次に、 ステップ S 6 8で、 上述 した出力直交変換行列となる基底の長さ L , の行列 T ( )を作成し、 次のステツプ S 6 9で、 この行列 T の低周波基底べク トル k個 を取り出して、 k行 L j 列から成る行列 Bを作成する。 最後のステ ップ S 7 0で、 上記式 ( 5 ) より、 変換行列 D j (二ひ jB A) を求 める。 ただし、 ひ」 は、 第 j次元における補正のためのスカラ一値 又はべク トル値である。
次に図 1 1は、 図 1 0の手順で作成された変換行列 Ε» ,· を用いて 各次元毎に解像度変換を行う場合の処理手順を示すフローチヤ一 ト である。
この図 1 1の最初のステツプ S 8 1では、 入力ディジ夕ル信号の 次元数の全ての次元で解像度変換が終了したか否かを判別しており、 全ての次元で変換していれば、 解像度変換処理を終了する。 次のス テツプ S 8 2では、 新しい次元 ( j ) を現在の次元に設定する。
次のステップ S 8 3では、 入力信号の大きさが L j の倍数に等し くないか否かを判別し、 等しくないときステップ S 8 4に進んで、 入力信号にダミ一デ一夕等を補足して L j の倍数になるようにして いる。 このように、 元々 L」 の倍数の、 あるいは補足により Lj の 倍数とされた入力信号は、 ステップ S 8 5により、 長さ L j のメタ プロックに分割される。
次のステップ S 8 6では、 現在の次元についての全メ夕ブロック を変換処理したか否かを判別し、 全メ夕ブロックの変換が終了して いるときには上記ステップ S 8 1に戻り、 全メタブロックの変換が 終了していないときにはステップ S 8 7に進んでいる。 ステップ S 8 7では、 次のメ夕ブロヅクを現在のメ夕ブロックべク トル Mに設 定し、 ステップ S 8 8で現在のメタブロックに上記変換行列 D j を 掛ける、 すなわち C = D ,,M を計算し、 ステップ S 8 9でこの Cを 出力して、 上記ステップ S 8 6に戻っている。
以上の図 1 0、 図 1 1に示した手順により、 n次元の入力ディ ジ タル信号に対して、 基底を変えずに、 第 j次元について解像度を 1 / m j に縮小することができる。 なお、 変換行列 D ., は、 図 1 0の 手順により一度求めておけば、 図 1 1の解像度変換処理中は反復し て使用できる。
以上説明したような本発明の第 1の実施の形態によれば、 D C T 等の直交変換により周波数領域に変換された入力ディジ夕ル信号に 対して、 変換基底を任意に変更することができ、 出力ディ ジタル信 号の基底の長さ及び全体の長さを調整することで、 任意の倍率で解 像度等を縮小することができる。 また、 出力ディ ジタル信号の直交 変換方式及び全体の長さを入力ディ ジ夕ル信号と同じにして本実施 の形態を適用することにより、 基底の長さが異なる出力ディジ夕ル 信号を得ることができる。 また、 出力ディジタル信号の直交変換方 式を入力ディ ジ夕ル信号と異なるものにして上記第 1の実施の形態 を適用することにより、 直交変換方式、 基底の長さ、 及び解像度の 異なる出力ディ ジタル信号を得ることができる。 また、 出力ディ ジ タル信号の直交変換方式及び全体の基底の長さを入力ディ ジ夕ル信 号と同じにして本実施の形態を適用することにより、 入力ディジ夕 ル信号に対する復号器をそのまま用いることができ、 解像度のみ変 換することができる。 また、 拡大と縮小とを組み合わせることによ り、 任意の倍率で解像度変換が可能である。 1次元のみならず、 複 数次元の周波数領域信号に適用することが可能である。 さらに、 任 意の倍率で変換可能であるため、 2次元信号における輝度信号や色 差信号等のように、 コンポーネン トによって変換ブロックの大きさ や構成が異なる場合でも、 コンポーネン ト間の対応を保ったまま解 像度変換が可能である。 またさらに、 変換に伴う計算が複雑でない ため、 計算誤差が生じにく く、 高品質の変換が可能であり、 変換に 伴う計算量が少ないため、 特別なハードウェアを必要とせず、 ソフ トウエアによって十分高速に変換できる。 さらに、 直交変換の後、 圧縮されたディ ジ夕ル信号に対しては、 圧縮の伸張のみ行うことで、 上記第 1の実施の形態により各種変換が可能になる。
ところで、 上述した本発明の第 1の実施の形態の説明においては、 処理の単位として、 いくつかのブロックをまとめたメタブロックに しており、 プロックの大きさを変えないで解像度変換が行えるのは、 倍率が 1 / n ( n = l , 2 , 3, · · · ) の場合であり、 また、 メ タブ口ックの一部分、 あるいはメタブ口ックの境界を跨った部分に ついて処理を行えないが、 これを次のように拡張することにより、 任意の倍率で解像度変換が行え、 プロックの境界を跨った任意の大 きさの部分信号についても処理が可能となる。
すなわち、 図 1 2は、 本発明の第 2の実施の形態となるディジ夕 ル信号変換方法及び変換行列作成方法を説明するための説明図であ る。
この図 1 2においては、 上記図 1の入力直交変換行列生成部 1 1 を拡張した連結行列生成部 1 1 1 と、 上記 [ 1 1の出力直交変換行列 生成部 1 2を拡張した分割行列生成部 1 1 2 と、 メタブロックの一 部分あるいはメ夕プロ ックの境界を跨った任意の部分を取り出すた めの選択行列を生成する選択行列生成部 1 1 7 と、 解像度変換処理 等の所望の信号処理を施すための加工行列を生成する加工行列生成 部 1 1 8と、 これらの 4つの行列から必要な行列を求める変換行列 生成部 1 1 3 とを備え、 この変換行列生成部 1 1 3からの変換行列 Dを信号変換部 1 1 4に送って、 直交変換された入力ディ ジタル信 号 1 1 5を、 直交変換された状態のままで変換処理して、 出力ディ ジ夕ル信号 1 1 6 としている。
この場合、 上記入力ディジタル信号 1 1 5は、 例えば図 1 3の ( a ) に示すように、 長さ k! のブロックが m! 個まとめられた長 さ ( = k i x m i ) のメタブロックの直交変換された信号 (例えば 周波数領域の信号) であり、 これが上記変換行列生成部 1 1 3から の変換行列 Dにより直交変換された信号のままで変換処理されて、 図 1 3の (b ) に示すように、 長さ k。 のプロックが m。 個まとめ られた長さ N ( = k。x m 0 ) のメタブロックの直交変換された信号 が上記出力ディ ジ夕ル信号 1 1 6 として取り出される。
次に、 上記図 1 2の連結行列生成部 1 1 1からの連結行列 A、 選 択行列生成部 1 1 7からの選択行列 W、 加工行列生成部 1 1 8から の加工行列 X、 分割行列生成部 1 1 2からの分割行列 Bのそれぞれ の機能について、 図 1 4を参照しながら説明する。 図 14において、 (a) は上記入力ディジ夕ル信号 1 1 5の上記 1メタブロック分に相当し、 この入力ディジタル信号 (a) は、 直 交変換された領域、 例えば周波数領域あるいは D C T等により圧縮 符号化された圧縮領域の信号である。 上記連結行列 Aは、 この圧縮 領域 (直交変換された領域) の長さ k, のブロックを 個まとめ て長さ L ( = k I X m のメタブロ ックに連結すると共に、 逆直交 変換を施して時間あるいは空間領域の信号 (b) に変換する機能を 有する。 上記選択行列 Wは、 この時間あるいは空間領域の信号 (b) から任意の処理の対象となる一部を取り出す機能を有する。 図 1 4 の例では、 上記長さ Lの時間あるいは空間領域の信 ¾· ( b ) の p番 目から q番目までの長さ Mの信号 (c) を取り出している。 上記加 ェ行列 Xは、 上述した解像度変換や縮小/拡大のような実際の信号 変換処理を上記時間あるいは空間領域で行うためのものであり、 ュ —ザが所望のものを指定すればよい。 図 14の例では、 長さ Mの信 号 ( c) を長さ Nの信号 (d) に変換している。 上記分割行列 Bは、 長さ N ( = k。xm。) の上記時間あるいは空間領域の信号 (d) を、 ブロック長 k。 の m。 個のブロックに分割すると共に直交変換して 直交変換された領域の信号 (e) としており、 この信号 ( e) が上 記出力ディジ夕ル信号 1 1 6に相当する。
次に、 上記各行列 A、 W、 X、 Bの生成についてさらに詳細に説 明する。
先ず、 上記連結行列生成部 1 1 1では、 上記入力ディ ジタル信号 1 1 5に対して予め施された直交変換を表す直交変換行列 Ts(kI) ' の逆行列 Ts(kI)1 を生成し、 この逆行列 Ts(kI) 1 を、 次の式 ( 7) に示すように対角上に 個並べて L次の正方行列、 すなわ ち L行 L列 (L kz Xim) の行列を作成し、 これを連結行列 Aと する。 この式 ( 7 ) は、 上記第 1の実施の形態の式 ( 2 ) に相当す るものである。
ん (7)
Figure imgf000027_0001
次に、 上記分割行列生成部 1 1 2では、 出力ディジタル信号に対 して施される逆直交変換を示す逆変換行列 Td(k0) 1 に対応する直 交変換行列 Td(k0) を生成し、 この行列 Td(k0) を、 次の式 ( 8 ) に示すように対角上に m。 個並べて、 N次の正方行列、 すなわち N 行 N列 (N = k。xm。) の行列を作成し、 これを分割行列 Bとする c
B
Figure imgf000027_0002
上記選択行列生成部 1 1 7では、 上述したようにメタブロックか ら処理の対象とすべき部分を選択するための選択行列 Wを生成する < この選択行列 Wは、 各列について 1である要素が 1つずつ有り、 残 りは 0という行列になる。 ここで一例として、 上記図 1 4の選択行 列 Wに示すように、 長さ Lのメタブロックから処理対象となる p番 目から q番目までの長さ Mの信号を取り出すための M行 L列の行列 Wを次の式 ( 9 ) に示す。 この式 ( 9 ) は、 L列の内の p列から q 列までの M列の部分が 1を対角上に配置した M次の正方行列 (M行 M列の行列) となっており、 残りの要素は全て 0となっている。
Figure imgf000028_0001
P M(=q-p) L-q 次に、 上記加工行列生成部 1 1 8では、 実際に信号の加工処理、 例えば解像度変換等を行うための加工行列 Xを生成する。 この加工 行列 Xについては、 直交変換された領域での処理を意識する必要は なく、 時間あるいは空間領域での原信号レベルで加工を行う行列を、 ユーザが任意に定めればよい。 この加工行列 Xは、 上記図 1 4の例 では N行 M列の行列となるが、 具体例については後述する。
次に、 上記図 1 2の変換行列生成部 1 1 3では、 上記各行列生成 部 1 1 1、 1 1 7、 1 1 8、 1 1 2で生成された行列を元に、 次の 式 ( 1 0 ) で計算されるような N行 L列の変換行列 D、 すなわち、
D = B ■ X · W · A · · · ( 1 0 ) を生成する。 ' 図 1 2の信号変換部 1 1 4では、 図 1 5に示すように、 入力信号 をメタブロック Mi ( i = 0 , 1, 2, · · ' ) に分割し、 各メタ ブロック Mi 毎に上記変換行列 Dを、
Ci =D · Mi · · · ( 1 1 ) のように適用していく。 なお、 図 1 5における各メタブロック M i の点線の間が、 選択行列 Wi によって選択される範囲である。 こう して得られた信号 C i を順に連結していく ことにより、 最終的な出 力信号を得る。 ここで、 信号 C i は、 上述した長さ Nのメタブロッ クであり、 長さ k。 のブロックの m。 個に分割されるものである。 ここで、 上記加工行列 Xは、 処理対象となる信号が直交変換され ていることを意識せずに、 時間あるいは空 f?;j領域の原信号について 処理を行うための行列を用いればよく、 種々の信号加工処理を用い ることができる。 例えば、 任意の倍率の解像度変換処理、 直交変換 の基底を変換する処理、 J P E Gや MP E G等の圧縮された画像に 対するエフェク ト処理ゃフィルタ リ ング処理、 ディ ジ夕ルビデオ信 号から MP E Gへ等のようなフォ一マツ ト変換処理、 等が挙げられ る。
次に、 具体例として、 上記入出力のブロック長 k i 、 k。 を変え ないで、 すなわち k t = k。二 kとして、 任 ;の倍率で信号の解像度 変換を行う場合について説明する。
先ず、 解像度変換の倍率ひを、 ひ = a/b (ただし、 a、 bは規 約) とする。 このとき、 L二 b k、 N= a kとして、 連結行列及び 分割行列を生成する。 また、 選択行列 Wは、 L =Mとし、 L次の単 位行列とする。 加工行列 Xとしては、 原信号について解像度変換を 行えるような行列であればどのようなものでもよいが、 ここでは一 例として、 次の式 ( 1 2 ) に示すような行列を用いるものとする。 この式 ( 1 2 ) 中の αは、 レベル補正のための係数である。
Figure imgf000030_0001
N L-N これらの行列を用いて、 上述したように変換行列 Dを生成し、 信 号変換処理を行うことにより、 原信号を復元することなく、 直交変 換された領域のままで任意の倍率で解像度変換を行うことができる < この具体例のような解像度変換を行う場合において、 上記図 1 2 の各行列生成部 1 1 1、 1 1 7、 1 1 8、 1 1 2における行列生成 の手順について、 図 1 6〜図 20のフ口一チヤ一トを参照しながら 説明する。 ここでは説明を簡略化するために、 上述したように、 入 カブロック長 と出力ブロック長 k。 とが同じ、 すなわち、 1^ = k0= kとしている。
図 1 6は、 上記連結行列生成部 1 1 1における連結行列 Aの生成 手順を説明するためのフローチヤ一トである。
この図 1 6において、 最初のステップ S 1 2 1では、 行列生成の ためのパラメ一夕を初期化する。 このパラメ一夕の初期化について は、 後で図 1 7を参照しながら説明する。 次のステップ S 1 22で は、 上記逆直交変換行列 Ts(k) 1を作成している。 次に、 ステツ'プ S 1 2 3で任意の L行 L列の行列 A, を生成し、 ステップ S 1 24 でこの行列 A, の要素を全て 0に初期化し、 ステップ S 1 25で 0 を要素とされた行列 A ' の対角上に上記逆直交変換行列 Ts(k)—1を mi 個配置して、 上述した連結行列 Aを生成する。 なお、 入カブ口 ック長 k ! と出力ブロック長 k。 とが異なる場合には、 上記ステツ プ S 1 2 2で作成される逆直交変換行列は T s 1 となることは 勿論である。
図 1 7は、 上記図 1 6のステップ S 1 2 1でのパラメ一夕の初期 化の手順の一例を示すフローチャートである。 この図 1 7のステツ プ S 1 6 1では、 入出力のプロック長を kとし、 ステップ S 1 6 2 及び S 1 6 3で上記解像度変換の倍率の分子 a及び分母 bをそれぞ れ設定し、 ステップ S 1 6 4で L = b x k、 ステップ S 1 6 5で N = a X kをそれぞれ設定している。 このパラメ一夕の初期化は、 以 下に説明する図 1 8〜図 2 0の各行列の生成手順においても同様に 用いられるものである。
図 1 8は、 上記選択行列生成部 1 1 7における選択行列 Wの生成 手順を説明するためのフ口一チャートである。 この図 1 8において、 最初のステップ S 1 3 1で上記図 1 7で説明したパラメ一夕の初期 化が行われ、 次のステップ S 1 3 2で L行 L列の単位行列 Iを生成 し、 次のステップ S 1 3 3で選択行列 Wをこの単位行列 I とする。 図 1 9は、 上記加工行列生成部 1 1 8における加工行列 Xの生成 手順を説明するためのフ口一チャートである。 この図 1 9において、 最初のステヅプ S 1 4 1で上記図 1 7で説明したパラメ一夕の初期 化を行い、 ステップ S 1 4 2で N行 N列の逆コサイン変換行列 I D C Tを求め、 ステップ S 1 4 3で L行 L列のコサイ ン変換行列 D C Tを求める。 次のステップ S 1 4 4では N行 L列の行列 Yを作成し、 ステップ S 1 4 5では解像度の倍率 o; = a / bを求め、 次のステツ プ S 1 4 6で上記式 ( 1 2 ) に相当する式により、 加工行列 Xを求 めている。 図 2 0は、 上記分割行列生成部 1 1 2における分割行列 Bの生成 手順を説明するためのフローチヤ一トである。 この図 2 0において、 最初のステツプ S 1 5 1で上記図 1 7で説明したパラメ一夕の初期 化を行い、 ステツプ S 1 5 2で上記逆直交変換行列 T d u 1を作成 し、 ステップ S 1 5 3で N行 N列の任意の行列 B ' を生成し、 ステ ップ S 1 5 4でこの行列 B ' の要素を全て 0に初期化し、 ステップ S 1 5 5でこの行列 B ' の対角上に上記逆直交変換行列 T d (k 1を mo 個配置して、 分割行列 Bを生成する。
次に、 解像度変換を行う別の手法として、 上記図 1 5に示したよ うに、 メタブロック M i ( i = 0, 1 , 2 , · · · ) から選択行列 を用いて必要な部分のみを抜き出して、 縮小/拡大を行うようにし てもよい。 この方法は、 前述の方法に比べて倍率の自由度が制限さ れるものの、 倍率の分子、 分母の値が大きい場合には、 変換行列の 大きさを小さく抑えることができるため、 よ り高速に変換を行うこ とができる。
次に、 この第 2の実施の形態についての多次元信号への適用につ いて説明する。 この場合、 加工行列 Xとして信号の次元に依存しな いようなものを選ぶことにより、 全体としての変換操作を各次元に ついて独立に行うことが可能となる。 従って、 この第 2の実施の形 態は、 多次元の信号処理に容易に拡張することが可能である。 例え ば、 画像のような 2次元について処理を行う場合には、 先ず縦方向 について処理した後、 続いて横方向に処理を行えばよい。 実際に、 上述した解像度変換の処理はいずれも多次元信号処理への拡張が可 能であり、 いわゆる J P E Gや M P E G等の圧縮画像情報の解像度 変換に用いることができる。 次に、 上記第 2の実施の形態のフォーマツ ト変換への応用につい て説明する。 フォーマッ トによっては、 様々な事情から部分的に変 則的な形式でデータが保存されている場合がある。 例えば、 いわゆ る D V (ディ ジタルビデオ) 規格のフォーマッ トにおいては、 図 2 1に示すように色差信号の右端で 4 x 8画素という半端なプロック ができるため、 上下 2つのブロックを組み合わせて仮想的に 8 8 のブロックを作って復号化するようにしている。
すなわち、 D V (ディジタルビデオ) 規格の色差信号は、 1 8 0 X 4 8 0の信号列となっていることから、 D C Tを施すために 8 8のマクロブロックに分割しょうとすると、 1 8 0 = ( 2 2 x 8 ) + 4 より、 水平方向に 4余り、 図 2 1に示すように、 右端に 4 X 8のブロックが生じてしまう。 そこで D V規格では、 色差信号を符 号化する際に、 図 2 1に示すように右端に生じた 4 X 8のブロック の上下で隣り合う 2プロックを組み合わせて 8 x 8のプロヅクを構 成し、 D C T符号化を施している。
従って、 D V規格の信号に解像度変換のような処理を施すために は、 図 2 2の ( a ) に示すように、 右端の 4 x 8のプロヅクの上下 に隣り合うプロック A、 Bを組み合わせて、 図 2 2の ( b ) に示す ように仮想的に 8 X 8のプロックを作って符号化した後、 図 2 2の ( c ) の斜線部、 すなわち本来その部分についての情報が配置され ている部分のみを処理対象として信号処理を行うことが必要とされ る。
こういった変則的なフォーマッ トについても、 上記選択行列 Wを 適切に選ぶことにより、 問題なく処理することができる。
例えば、 上述のような D V規格の色差信号を処理する場合には、 次の式 - o o o o o( 1 3) で示す選択行列 W, と、 次の式 ( 14) で示す選択 行列 W2 とを、 ブロックの y座標の値に応じて使い分けることによ り、 図 22の ( c) の斜線部に対する処理を有効に行うことができ る o
0 1 0 0 0 0 0 0 0 0 0 0、
0 0 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
(1 3)
0 0 1 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0;
Figure imgf000034_0001
次に、 上記第 2の実施の形態の動画像編集への応用について説明 する。
例えばいわゆる MP E Gのような圧縮画像の編集を行う際、 力ヅ 卜の切換に際して、 例えば前の力ッ トをパンァゥ トしながら後の力 ッ トを重ねてつなげる等のエフェク 卜機能が必要となる。 このよう な処理を実現するためには、 直交変換された画像の一部分を抜き出 す等の処理が必要となるが、 上記第 2の実施の形態を用いることに より、 原画像を復元することなく、 プロックの境界に制限されずに 任意の部分を抽出し加工することが可能となる。
次に図 23は、 上述したような処理を実現するためのハードゥエ ァの概略構成の一例として、 特にディジ夕ル画像信号を取り扱う場 合のブロック図を示している。 この図 23において、 デ一夕計算や 信号処理のための CPU (中央演算処理ュニッ ト) 10 1が接続さ れたバスには、 プログラムやデータが記憶された ROM 1 02、 読 出書込可能な記憶手段である RAM、 画像人カイ ンターフ ェース 1 04、 画像出力インターフェース 1 06、 ハードディスク等の記録 媒体 1 08が接続されており、 端子 105より画像入カイン夕一フ エース 1 04を介してディジ夕ル画像信号が入力され、 画像出カイ ン夕ーフエース 106より端子 107を介してディ ジ夕ル画像信号 が出力される。
なお、 本発明の実施の形態の処理に用いるプログラムは、 ROM 102又はハー ドディスク等の記録媒体 1 0 8に記録され、 提供さ れる。 また、 CD— ROM等のディスクや、 通信回線を介して提供 することも可能である。
次に、 上述した本発明の実施の形態を、 いわゆる DV規格から M P E G 1規格に変換するシステムに適用した例について、 図 24を 参照しながら説明する。
この図 24に示すシステムにおいては、 先ず、 家庭用の DV規格 のディ ジ夕ルビデオカメラ 20 1で撮影された映像情報を、 I E E E 1394イン夕一フエースを介して、 ディ ジ夕ル信号のまま直接 にパーソナルコンピュータ 2 0 2内に取り込む。 ここで、 D V規格 の信号はデータ量が膨大であり、 パーソナルコンピュータ 2 0 2内 で取り扱うには適切でない。 そこで、 上記本発明の実施の形態を用 いて、 この D V規格の映像データを M P E G 1規格の映像データに 変換する。 こうすることによりデ一夕量が大幅に小さくなり、 パー ソナルコンビュ一夕 2 0 2内でのソフ トウエアによるノンリニア編 集を行ったり、 映像をいわゆるィン夕一ネッ ト 2 0 3を通じて他の 機器 2 0 4に提供する等が可能となる。
上記パーソナルコンビュ一夕 2 0 2内での D V規格から M P E G 1規格への方式変換は、 図 2 5の変換処理部 2 1 0により実現され る。
この図 2 5において、 供給された D V映像デ一夕は、 変換処理部 2 1 0内の可変長復号化部 2 1 1で可変長復号化され、 逆量子化部 2 1 2で逆量子化されて、 直交変換された領域あるいは圧縮領域の ままで解像度変換部 2 1 3に送られ、 解像度変換がなされる。 この 解像度変換部 2 1 3での変換処理が上述した実施の形態により実現 される。 また、 動き予測部 2 1 4において直交変換された領域ある いは圧縮領域のままで動き予測処理が施された後、 量子化部 2 1 5 に送られて量子化され、 可変長符号化部 2 1 6で可変長符号化され て、 M P E G 1規格の映像デ一夕となって取り出される。
ここで、 従来の方式変換によれば、 上記逆量子化部 2 1 2で逆量 子化された信号を、 逆 D C T部 2 2 1で時間あるいは空間領域の信 号に逆直交変換した後、 デ一夕形式変換部 2 2 2でデータ形式の変 換を行い、 解像度変換部 2 2 3で解像度変換を、 動き予測部 2 2 4 で動き予測を順次行う。 次に、 D C T部 2 2 5で直交変換を行って 直交変換された領域あるいは圧縮領域に変換した後、 上記量子化部
2 1 5に送ることが必要とされる。 このように、 従来の手法では、 直交変換された領域あるいは圧縮領域の信号を、 時間あるいは空間 領域の信号に逆直交変換した後に、 必要な信^処理を行って、 再び 直交変換する手間がかかるため、 処理が複雑化し、 処理速度が低下 するが、 本発明の実施の形態を用いることにより、 大幅に処理ステ ップを削減でき、 ソフ トウェアによる高速処理が可能となる。
以上説明した本発明の第 2の実施の形態によれば、 原信号を復元 する必要がないので高速に処理することができ、 計算誤差に伴う信 号の劣化が少ないのみならず、 任意の倍率で解像度変換を行うこと ができ、 信号のブロックサイズを任意の大きさに変換することがで きる。 また、 選択行列により信号の一部だけを選択的に処理するこ とができ、 その際、 入力信号のブロックの境界を意識する必要がな い。 また、 加工行列については、 信号の I :交変換基底に影饗されず に決めることができ、 行列のデザイ ンが容易で、 様々な信号処理に 適用することができる。
以上の説明からも明らかなように、 本発明によれば、 逆直交変換 のための逆直交変換行列 T s 1 及び直交変換のための直交変換行列 T d を生成し、 生成された上記逆直交変換行列 T s—1 及び直交変換 行列 T d を用いて変換行列 Dを生成し、 直交変換された領域の入力 ディジ夕ル信号に対して上記変換行列 Dを用いて直交変換された領 域で変換処理を行い、 直交変換された領域の出力ディジ夕ル信号を 得ているため、 直交変換された領域で解像度や変換基底の変換が直 接的に行え、 時間領域や空間領域への復号 (逆直交変換) が不要と なって、 計算が簡略化され、 計算誤差の小さい高品質の変換が行え ると共に、 ハードウェア負担も軽減され、 ソフ トウェアによって十 分高速に変換処理が実現できる。
上記入力ディ ジ夕ル信号は、 時間領域や空間領域の原信号を上記 逆直交変換行列 T s—1 に対応する直交変換行列 T S により直交変換 して得られたディジ夕ル信号であり、 上記出力ディ ジ夕ル信号は、 上記直交変換行列 T d に対応する逆直交変換行列 T cT 1 により復号 されて時間領域や空間領域の信号に変換されるものとすることが挙 げられ、 また、 上記原信号を長さ kの変換ブロック毎に基底の長さ kの上記直交変換行列 T s により直交変換して上記入力デイジ夕 ル信号とし、 この入力ディ ジ夕ル信号の隣接する m個のプロックか ら成る長さ L ( = k x m ) の連続する信号を、 長さ Nのブロック 1 個に変換するようにすればよく、 上記変換行列 Dは、 上記直交変換 行列 T s ( k )を m個対角軸上に配置して L次の正方行列 Aを作成し、 基底の長さ Lの直交変換行列 T d (いの低周波基底べク トル N個を取 り出して N行 L列の行列 Bを作成し、
Ό = a · · A
ただし、 ひはスカラ一値又はべク トル値のレベル補正のための 係数
の式により作成することができる。
これにより、 直交変換された領域で、 入力ディ ジタル信号を N / Lに縮小変換することができる。 特に N = Lのとき、 変換行列 Dは、 変換基底を変換する行列になる。 また、 直交変換行列 T s と T d と が異なる場合には、 直交変換方式の変換を行うことができる。
また、 上記入力ディジ夕ル信号として n次元のディジ夕ル信号を 用い、 各次元毎に上記変換行列 Dを用いて直交変換された領域で変 換処理を行い、 n次元の出力ディジ夕ル信号を得ることができる。 このとき、 各次元毎に変換倍率が異なる場合には、 各次元毎の変換 行列を作成してこれらの変換行列を用いて各次元毎に異なる倍率で 変換処理を行わせることで、 輝度信号や色差信号等のコンポーネン 卜によって変換ブロックの大きさや構成が異なる場合でも、 コンポ ーネン ト間の対応を保ったまま解像度変換を行うことができる。 さらに、 直交変換された入力ディ ジ夕ル信号の変換プロックをそ のまま用い、 所定の値を埋めることにより長さが m倍のプロックに 変換し、 直交変換された領域の出力ディジ夕ル信号を得ることによ り、 解像度の拡大処理を直交変換された領域で行うことができ、 こ の拡大と上記縮小とを組み合わせることにより、 任意の倍率で解像 度変換が行える。
また、 直交変換された領域の入力ディジ夕ル信号の一定サンプル 数のブロックを少なく とも 1つ連結してメ夕ブロックを形成すると 共に逆直交変換する連結行列を生成し、 上記メ夕プロックから処理 の対象とすべき部分を選択して取り出す選択行列を生成し、 上記選 択された部分に対して所望の信号加工処理を施す加工行列を生成し、 上記信号加工処理された信号に対して直交変換を施すと共に少なく とも 1つのブロックに分割する機能を有する分割行列を生成し、 上 記連結行列、 選択行列、 加工行列、 分割行列を用いて直交変換され た領域の信号に対して変換処理を施すための変換行列を生成し、 直 交変換された入力ディ ジ夕ル信号に対して上記メ夕プロック毎に上 記変換行列を適用して出力ディ ジタル信号を得るようにしているた め、 原信号を復元する必要がないので、 高速に処理することができ、 解散誤差に伴う信号の劣化が少ないのみならず、 任意の倍率で解像 度変換を行うことができ、 信号のブロックサイズを任意の大きさに 変換することができる。 また、 選択行列により信号の一部だけを選 択的に処理することができ、 その際、 入力信号のブロックの境界を 意識する必要がない。 また、 加工行列については、 信号の直交変換 基底に影響されずに決めることができ、 行列のデザインが容易で、 様々な信号処理に適用することができる。
なお、 本発明は上述した実施の形態のみに限定されるものではな く、 例えば、 上記実施の形態においては主として直交変換された画 像信号の解像度変換について説明したが、 この他、 直交変換された オーディォ信号等の種々の周波数領域のディ ジ夕ル信号に対して本 発明を適用できることは勿論である。 また、 直交変換された信号は 周波数領域の信号に限定されず、 原信号は時間領域あるいは空間領 域等の信号が用いられる。

Claims

請 求 の 範 囲
1. 逆直交変換のための逆直交変換行列 T s 1 及び直交変換のた めの直交変換行列 T d を生成するェ程と、
生成された上記逆直交変換行列 Ts ' 及び I 交変換行列 Td を用 いて変換行列 Dを生成する工程と、
直交変換された領域の入力ディジ夕ル信 に対して上記変換行列 Dを用いて直交変換された領域で信号処理を行い出力ディ ジタル信 号を得る工程と
を有するディ ジ夕ル信号変換方法。
2. 上記入力ディジタル信号は、 時間領域の原信号を上記逆直交 変換行列 Ts— 1 に対応する直交変換行列 Ts により直交変換して得 られたディ ジ夕ル信号であり、
上記出力ディ ジタル信号は、 上記直交変換行列 Td に対応する逆 直交変換行列 Td— 1 により復号されて時間領域の信号に変換される ものである
請求の範囲第 1項記載のディ ジ夕ル信号変換方法。
3. 上記原信号を長さ kの変換プロック毎に基底の長さ kの上記 直交変換行列 T s d により直交変換した信号を上記入力デイ ジタル 信号とし、 この入力ディジ夕ル信号の隣接する m個のプロックから 成る長さ L ( = k xm) の連続する信号を、 上記変換行列 Dを用い て、 長さ Nのブロック 1個に変換する
請求の範囲第 2項記載のディ ジ夕ル信号変換方法。
4. 上記直交変換行列 Ts )に対する逆直交変換行列 Ts(k) 1を m個対角軸上に配置して L次の正方行列 Aを作成し、 基底の長さ L の直交変換行列 T d (いの低周波基底べク トル N個を取り出して N行 L列の行列 Bを作成し、
D =ひ · B · A
ただし、 ひはスカラ一値又はベク トル値
の式により上記変換行列 Dを作成する
請求の範囲第 3項記載のディジ夕ル信号変換方法。
5 . 上記ブロック長 Nは、 上記長さ Lに等しい請求の範囲第 3項 記載のディジ夕ル信号変換方法。
6 . 上記入力ディ ジ夕ル信号は n次元のディ ジ夕ル信号であり、 各次元毎に上記変換行列 Dを用いて直交変換された領域で信号処 理を行い、 n次元の出力ディ ジタル信号を得る
請求の範囲第 1項記載のディジ夕ル信号変換方法。
7 . 上記入力ディ ジタル信号は n次元のデイ ジタル信号であり、 各次元毎に異なる倍率で変換を行うための変換行列を作成し、 これらの変換行列を用いて各次元毎に異なる倍率で変換処理を行 い、 n次元の出力ディ ジタル信号を得る
請求の範囲第 1項記載のディジ夕ル信号変換方法。
8 . 直交変換された入力ディジ夕ル信号の変換プロックをそのま ま用いて所定の値を補うことにより長さが m倍のプロックに変換し、 周波数領域の出力ディ ジ夕ル信号を得る
請求の範囲第 1項記載のディジ夕ル信号変換方法。
9 . 逆直交変換のための逆直交変換行列 T s—1 及び直交変換のた めの直交変換行列 T d を生成する直交変換行列生成手段と、
生成された逆直交変換行列 T s—1 及び直交変換行列 T d を用いて 変換行列 Dを生成する変換行列生成手段と、
直交変換された領域の入力ディジ夕ル信号に対して上記変換行列
Dを用いて直交変換された領域で信号処理を行い出力ディ ジ夕ル信 号を得る信号変換手段と
を有するディ ジタル信号変換装置。
1 0. 上記入力ディ ジタル信号は、 時間領域の原信号を上記逆直 交変換行列 Ts ' に対応する直交変換行列 Ts により直交変換して 得られたディ ジ夕ル信号であり、
上記出力ディ ジタル信号は、 上記直交変換行列 Td に対応する逆 直交変換行列 Td 1 により復号されて時間領域の信号に変換される ものである
請求の範囲第 9項記載のディジ夕ル信号変換装置。
1 1. 上記原信号を長さ kの変換ブロック毎に基底の長さ kの上 記直交変換行列 Ts(k)により直交変換した信号を上記入力ディ ジ夕 ル信号とし、
上記信号変換手段は、
上記入力ディ ジ夕ル信号の隣接する m個のプロックから成る長さ L ( = kxm) の連続する信号を、 上記変換行列 Dを用いて、 長さ Nのプロック 1個に変換する
請求の範囲第 1 0項記載のディジ夕ル信号変換装置。
1 2. 上記変換行列生成手段は、
上記直交変換行列 Ts(k)に対する逆直交変換行列 Ts ) 1を m個 対角軸上に配置して L次の正方行列 Aを作成し、 基底の長さ Lの直 交変換行列 Td の低周波基底べク トル N個を取り出して N行 L列 の行列 Bを作成し、 D = ひ · B · A
ただし、 ひはスカラー値又はベク トル値
の式により上記変換行列 Dを作成する
請求の範囲第 1 1項記載のディジタル信号変換装置。
1 3 . 上記ブロ ック長 Nは、 上記長さ Lに等しい詰求の範囲第 1 1項記載のディ ジ夕ル信号変換方法。
1 4 . 上記入力ディ ジ夕ル信号として n次元のディジ夕ル信号を 入力し、
上記信号変換手段は、
各次元毎に上記変換行列 Dを用いて直交変換された領域で信号処 理を行い、 n次元の出力ディジタル信号を得る
請求の範囲第 9項記載のディジ夕ル信号変換装置。
1 5 . 上記入力ディ ジ夕ル信号は n次元のディジ夕ル信号であり、 上記変換行列生成手段は、
各次元毎に異なる倍率で変換を行うための変換行列を作成し、 上記信号変換手段は、
これらの変換行列を用いて各次元毎に異なる倍率で変換処理を行 い、 n次元の出力ディ ジタル信号を得る
請求の範囲第 9項記載のディジ夕ル信号変換装置。
1 6 . 上記信号変換手段は、
直交変換された入力ディジ夕ル信号の変換プロックをそのまま用 いて所定の値を補うことにより長さが m倍のプロックに変換し、 周 波数領域の出力ディ ジ夕ル信号を得る
請求の範囲第 9項記載のディジ夕ル信号変換装置。
1 7 . 逆直交変換のための逆直交変換行列 T s—1 及び直交変換の ための直交変換行列 Td を生成する工程と、
生成された上記逆直交変換行列 Ts—1 及び直交変換行列 Td を用 いて変換行列 Dを生成する工程と、
直交変換された領域の入力ディ ジ夕ル信号に対して上記変換行列 Dを用いて直交変換された領域で信号処 J を行 I、出力ディ ジタル信 号を得る工程と
を有する処理情報を提供する提供媒体。
1 8. 上記入力ディ ジタル信号は、 時間領域の原信号を上記逆直 交変換行列 Ts— 1 に対応する直交変換行列 Ts によ り直交変換して 得られたディジ夕ル信号であり、
上記出力ディ ジタル信号は、 上記直交変換行列 Td に対応する逆 直交変換行列 Td—1 により復号されて時間領域の信号に変換される ものである
請求の範囲第 1 7項記載の提供媒体。
1 9. 周波数領域の入力ディ ジタル信号に施された直交変換を表 す直交変換行列 Ts に対する逆直交変換行列 Ts 1 を生成する入力 直交変換行列生成工程と、
周波数領域の出力ディ ジ夕ル信号に対して施される復号処理とし ての逆直交変換を表す逆直交変換行列 Td—1 に対する直交変換行列 Td を生成する出力直交変換行列生成工程と、
上記逆直交変換行列 Ts—1 及び上記直交変換行列 Td に基づいて 変換行列 Dを生成する変換行列生成工程と、
を有する変換行列作成方法。
2 0. 周波数領域の入力ディジ夕ル信号に施された直交変換を表 す直交変換行列 Ts に対する逆直交変換行列 Ts—1 を生成する入力 直交変換行列生成手段と、
周波数領域の出力ディジ夕ル信号に対して施される復号処理とし ての逆直交変換を表す逆直交変換行列 T cT 1 に対する直交変換行列
T d を生成する出力直交変換行列生成手段と、
上記逆直交変換行列 T s 1 及び 記直交変換行列 T d に基づいて 変換行列 Dを生成する変換行列生成手段と、
を有する変換行列作成装置。
2 1 . 直交変換された領域の入力ディジタル信号の一定サンプル 数からなるブロックを少なく とも 1つ連結してメ夕プロヅクを形成 すると共に逆直交変換する連結行列を生成する Γ.程と、
上記メ夕ブ口ックから処理の対象とすべき部分を選択して取り出 す選択行列を生成する工程と、
上記選択された部分に対して所望の信号加工処理を施す加工行列 を生成する工程と、
上記信号加工処理された信号に対して直交変換を施すと共に少な く とも 1つのプロックに分割する機能を有する分割行列を生成する 工程と、
上記連結行列、 選択行列、 加工行列、 分割行列を用いて直交変換 された領域の信号に対して変換処理を施すための変換行列を生成す る工程と、
直交変換された入力ディジ夕ル信号に対して上記メ夕ブロック毎 に上記変換行列を適用して出力ディ ジ夕ル信号を得る工程と を有するディ ジ夕ル信号変換方法。
2 2 . 上記入力ディ ジ夕ル信号は、 時間領域の原信号を直交変換 行列 T s により直交変換して得られたディジ夕ル信号であり、 上記メタブロックは、 長さ のブロックを 個連結して成り、 上記連結行列は、 上記直交変換行列 T s に対応する逆直交変換行 列 T s— 1 を対角上に 個配置して成る
請求の範囲第 2 1項記載のディジタル信号変換方法。
2 3 . 上記出力ディ ジタル信号は、 逆直交変換行列 T d 1 により 復号されて時間領域の信号に変換される信号であり、
上記入力ディ ジ夕ル信号の 1つのメタプロックに変換されて得ら れる出力ディジ夕ル信号は、 長さ k。 のプロックを m。 個連結され て成り、
上記分割行列は、 上記直交変換行列 T d に対応する逆直交変換行 列 T d— 1 を対角上に m。 個配置して成る
請求の範囲第 2 1項記載のディジ夕ル信号変換方法。
2 4 . 直交変換された領域の入力ディジ夕ル信号の一定サンプル 数のブロックを少なく とも 1つ連結してメ夕ブ口ックを形成すると 共に逆直交変換する連結行列を生成する連結行列生成手段と、 上記メ夕プロックから処理の対象とすべき部分を選択して取り出 す選択行列を生成する選択行列生成手段と、
上記選択された部分に対して所望の信号加工処理を施す加工行列 を生成する加工行列生成手段と、
上記信号加工処理された信号に対して直交変換を施すと共に少な く とも 1つのプロックに分割する機能を有する分割行列を生成する 分割行列生成手段と、
上記連結行列、 選択行列、 加工行列、 分割行列を用いて直交変換 された領域の信号に対して変換処理を施すための変換行列を生成す る変換行列生成手段と、 直交変換された入力ディ ジ夕ル信号に対して上記メ夕プロック毎 に上記変換行列を適用して出力ディジ夕ル信号を得る信号変換手段 と
を有するディ ジ夕ル信号変換装置。
2 5 . 上記入力ディ ジタル信号は、 時間領域の原信号を直交変換 行列 T s により直交変換して得られたディ ジ夕ル信号であり、 上記メタプロックは、 長さ のブロックを 個連結して成り、 上記連結行列は、 上記直交変換行列 T s に対応する逆直交変換行 列 T s— 1 を対角上に 個配置して成る
請求の範囲第 2 4項記載のディ ジタル信号変換装置。
2 6 . 上記出力ディ ジタル信号は、 逆直交変換行列 T cT 1 によ り 復号されて時間領域の信号に変換される信号であり、
上記入力ディ ジ夕ル信号の 1つのメタブロックに変換されて得ら れる出力ディジタル信号は、 長さ k。 のブロ ックを m。 個連結され て成り、
上記分割行列は、 上記直交変換行列 T d に対応する逆直交変換行 列 T cT 1 を対角上に m。 個配置して成る
請求の範囲第 2 4項記載のディジ夕ル信号変換装置。
2 7 . 直交変換された領域の入力ディジ夕ル信号の一定サンプル 数からなるブロックを少なく とも 1つ連結してメタブロックを形成 すると共に逆直交変換する連結行列を生成する工程と、
上記メタブロックから処理の対象とすべき部分を選択して取り出 す選択行列を生成する工程と、
上記選択された部分に対して所望の信号加工処理を施す加工行列 を生成する工程と、 上記信号加工処理された信号に対して直交変換を施すと共に少な く とも 1つのブロックに分割する機能を有する分割行列を生成する 工程と、
上記連結行列、 選択行列、 加工行列、 分割行列を用いて直交変換 された領域の信号に対して変換処理を施すための変換行列を生成す る工程と、
直交変換された入力ディ ジ夕ル信号に対して上記メ夕プロック毎 に上記変換行列を適用して出力ディ ジ夕ル信^を得る信号変換工程 と
を有する処理情報を提供する提供媒体。
2 8 . 直交変換された領域の入力ディ ジ夕ル信号の一定サンプル 数のブロックを少なく とも 1つ連結してメ夕プロックを形成すると 共に逆直交変換する連結行列を生成する工程と、
上記メタブロ ックから処理の対象とすべき部分を選択して取り出 す選択行列を生成する工程と、
上記選択された部分に対して所望の信号加工処理を施す加工行列 を生成する工程と、
上記信号加工処理された信号に対して直交変換を施すと共に少な く とも 1つのプロックに分割する機能を有する分割行列を生成する 工程と、
上記連結行列、 選択行列、 加工行列、 分割行列を用いて直交変換 された領域の信号に対して変換処理を施すための変換行列を生成す る工程と
を有する変換行列生成方法。
2 9 . 直交変換された領域の入力ディジ夕ル信号の一定サンプル 数のプロックを少なく とも 1つ連結してメ夕プロックを形成すると 共に逆直交変換する連結行列を生成する連結行列生成手段と、 上記メ夕プロックから処理の対象とすべき部分を選択して取り出 す選択行列を生成する選択行列生成手段と、
上記選択された部分に対して所望の信号 11 X処理を施す加工.行列 を生成する加工行列生成手段と、
上記信号加ェ処理された信号に対して直交変換を施すと共に少な く とも 1つのブロックに分割する機能を有する分割行列を生成する 分割行列生成手段と、
上記連結行列、 選択行列、 加工行列、 分割行列を用いて直交変換 された領域の信号に対して変換処理を施すための変換行列を生成す る変換行列生成手段と
を有する変換行列生成装置。
補正書の請求の範囲
[ 1 9 9 8年 1 1月 1 6日 (1 6 . 1 1 . 9 8 ) 国際事務局受理:新しい請求の範囲 0 - 3 5が加えられた;他の請求の範囲は変更なし。 ( 5頁) ]
数のプロックを少なく とも 1つ連結してメ夕プロックを形成すると 共に逆直交変換する連結行列を生成する連結行列生成手段と、 上記メ夕プロックから処理の対象とすべき部分を選択して取り出 す選択行列を生成する選択行列生成手段と、
上記選択された部分に対して所望の信号加工処理を施す加工行列 を生成する加工行列生成手段と、
上記信号加工処理された信号に対して直交変換を施すと共に少な く とも 1つのプロ ックに分割する機能を有する分割行列を生成する 分割行列生成手段と、
上記連結行列、 選択行列、 加工行列、 分割行列を用いて直交変換 された領域の信号に対して変換処理を施すための変換行列を生成す る変換行列生成手段と
を有する変換行列生成装置。
3 0 . (追加) ディジ夕ル信号に所定の変換処理を施すディ ジ夕 ル信号変換方法において、
直交変換された領域のディジ夕ル信号を入力する工程と、 上記入力されたディジ夕ル信号に対して、 直交変換された領域で 変換行列 Dを用いて演算処理を行う工程と、
上記演算結果のディジタル信号を出力する工程と を有し、
上記変換行列 Dは、 逆直交変換のための逆直交変換行列 T s—1及び 直交変換のための直交変換行列 T dを用いて生成された行列である ディジ夕ル信号変換方法。
3 1 . (追加) ディジ夕ル信号に所定の変換処理を施すディ ジ夕 ル信号変換装置において、
補正された用紙 (条約第 19条) 直交変換された領域のディジ夕ル信号を入力する入力手段と、 上記入力されたディジ夕ル信号に対して、 直交変換された領域で 変換行列 Dを用いて演算処理を行う演算手段と、
上記演算結果のデイ ジ夕ル信号を出力する出力手段と
を有し、
上記変換行列 Dは、 逆直交変換のための逆直交変換行列 T s 1及び 直交変換のための直交変換行列 T dを用いて生成された行列である ディジ夕ル信号変換装置。
3 2 . (追加) ディジタル信号に所定の変換処理を施すための処 理命令を提供する提供媒体において、
直交変換された領域のデイ ジタル信号を入力する処理命令と、 上記入力されたディジ夕ル信号に対して、 直交変換された領域で 変換行列 Dを用いて演算処理を行う処理命令と、
上記演算結果のディジ夕ル信号を出力する処理命令と
を有し、
上記変換行列 Dは、 逆直交変換のための逆直交変換行列 T s 1及び 直交変換のための直交変換行列 T dを用いて生成された行列である 提供媒体。
3 3 . (追加) ディジ夕ル信号に所定の変換処理を施すディ ジ夕 ル信号変換方法において、
直交変換された領域のディジ夕ル信号を、 一定サンプル数からな るブロックを単位として入力する工程と、
上記入力されたディジタル信号に対して、 上記少なく とも 1つの - ブロックからなるメタプロック毎に、 直交変換された領域で変換行 列を用いて演算処理を行う工程と、
補正された用紙 (条約第 19条) 上記演算結果のディジタル信号を出力する工程と
を有し、
上記変換行列は、
上記プロヅクを少なく とも 1つ連結してメ夕プロックを形成する と共に逆直交変換する連結行列と、
上記メ夕プロックから処理の対象とすべき部分を選択して取り出 す選択行列と、
上記選択された部分に対して所望の信号加工処理を施す加工行列 と、
上記信号加工処理された信号に対して直交変換を施すと共に少な く とも 1つのプロックに分割する分割行列と
を用いて生成された行列である
ディ ジ夕ル信号変換方法。
3 4 . (追加) ディジ夕ル信号に所定の変換処理を施すディ ジタ ル信号変換装置において、
直交変換された領域のディ ジ夕ル信号を、 一定サンプル数からな るブロックを単位として入力する入力手段と、
上記入力されたディジ夕ル信号に対して、 上記少なく とも 1つの プロックからなるメタプロック毎に、 直交変換された領域で変換行 列を用いて演算処理を行う演算手段と、
上記演算結果のディジ夕ル信号を出力する出力手段と
を有し、
上記変換行列は、 - 上記プロックを少なく とも 1つ連結してメタプロックを形成する と共に逆直交変換する連結行列と、
捕正された用紙 (条約第 19条) 上記メ夕プロックから処理の対象とすべき部分を選択して取り出 す選択行列と、
上記選択された部分に対して所望の信号加工処理を施す加工行列 と、
上記信号加工処理された信号に対して直交変換を施すと共に少な く とも 1つのブロックに分割する分割行列と
を用いて生成された行列である
ディ ジタル信号変換装置。
3 5 . (追加) ディジタル信号に所定の変換処理を施すための処 理命令を提供する提供媒体において、
直交変換された領域のディジ夕ル信号を、 一定サンプル数からな るプロックを単位として入力する処理命令と、
上記入力されたディジ夕ル信号に対して、 上記少なく とも 1つの ブロックからなるメタプロック毎に、 直交変換された領域で変換行 列を用いて演算処理を行う処理命令と、
上記演算結果のディジ夕ル信号を出力する処理命令と
を有し、
上記変換行列は、
上記プロックを少なく とも 1つ連結してメ夕プロックを形成する と共に逆直交変換する連結行列と、
上記メ夕プロックから処理の対象とすべき部分を選択して取り出 す選択行列と、
上記選択された部分に対して所望の信号加工処理を施す加工行列 と、
上記信号加工処理された信号に対して直交変換を施すと共に少な
補正された用紙 (条約第 19条) く とも 1つのブロックに分割する分割行列と を用いて生成された行列である
提供媒体。
補正された用紙 (条約第 19条)
PCT/JP1998/002653 1997-06-16 1998-06-16 Procede et dispositif de conversion de signal numerique, dispositif et procede de generation de matrice de transformation et support d'alimentation Ceased WO1998058328A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP19980924648 EP0932106A1 (en) 1997-06-16 1998-06-16 Digital signal converting method and device, transformation matrix generating method and device, and supply medium
CA 2263454 CA2263454A1 (en) 1997-06-16 1998-06-16 Digital signal converting method and device, transformation matrix generating method and device, and supply medium

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
JP15904097 1997-06-16
JP9/159040 1997-06-16
JP9/238678 1997-09-03
JP9238678A JPH1173410A (ja) 1997-06-16 1997-09-03 ディジタル信号変換方法及び装置並びに変換行列作成方法

Publications (1)

Publication Number Publication Date
WO1998058328A1 true WO1998058328A1 (fr) 1998-12-23

Family

ID=26485961

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP1998/002653 Ceased WO1998058328A1 (fr) 1997-06-16 1998-06-16 Procede et dispositif de conversion de signal numerique, dispositif et procede de generation de matrice de transformation et support d'alimentation

Country Status (6)

Country Link
EP (1) EP0932106A1 (ja)
JP (1) JPH1173410A (ja)
KR (1) KR20000068164A (ja)
CN (1) CN1236456A (ja)
CA (1) CA2263454A1 (ja)
WO (1) WO1998058328A1 (ja)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2346324C2 (ru) * 2006-12-20 2009-02-10 Николай Александрович Емельяненко Устройство цифровой обработки полифазных ортогональных фазокодоманипулированных сигналов
US7630439B2 (en) 2003-06-18 2009-12-08 Kddi Corporation Apparatus for converting conversion encoding coefficients using sampling/resolution conversion
WO2011009196A1 (en) * 2009-07-20 2011-01-27 Valorbec, Societe En Commandite Finite dataset interpolation method

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1999023834A1 (fr) 1997-11-05 1999-05-14 Sony Corporation Procede de conversion de signaux numeriques et appareil de conversion de signaux numeriques
EP1211896A3 (en) * 2000-11-22 2006-04-19 Matsushita Electric Industrial Co., Ltd. System and method of providing data in one of a plurality of digital formats using an active disk
US8447121B2 (en) * 2005-09-14 2013-05-21 Microsoft Corporation Efficient integrated digital video transcoding
JP2013098730A (ja) * 2011-10-31 2013-05-20 Toshiba Corp 解像度変換装置、解像度変換方法及び動画像復号化装置
JP2015061143A (ja) * 2013-09-18 2015-03-30 沖電気工業株式会社 映像符号化装置、および、映像符号化プログラム

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0447860A (ja) * 1990-06-15 1992-02-18 Nippon Telegr & Teleph Corp <Ntt> 解像度変換処理方法
JPH04229382A (ja) * 1990-12-27 1992-08-18 Ricoh Co Ltd ディジタル画像データの解像度交換装置
JPH05316357A (ja) * 1992-05-14 1993-11-26 Nippon Telegr & Teleph Corp <Ntt> 解像度変換方法
JPH0898173A (ja) * 1994-09-21 1996-04-12 Hitachi Ltd 解像度変換装置およびそれを用いた解像度変換システム

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0447860A (ja) * 1990-06-15 1992-02-18 Nippon Telegr & Teleph Corp <Ntt> 解像度変換処理方法
JPH04229382A (ja) * 1990-12-27 1992-08-18 Ricoh Co Ltd ディジタル画像データの解像度交換装置
JPH05316357A (ja) * 1992-05-14 1993-11-26 Nippon Telegr & Teleph Corp <Ntt> 解像度変換方法
JPH0898173A (ja) * 1994-09-21 1996-04-12 Hitachi Ltd 解像度変換装置およびそれを用いた解像度変換システム

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7630439B2 (en) 2003-06-18 2009-12-08 Kddi Corporation Apparatus for converting conversion encoding coefficients using sampling/resolution conversion
RU2346324C2 (ru) * 2006-12-20 2009-02-10 Николай Александрович Емельяненко Устройство цифровой обработки полифазных ортогональных фазокодоманипулированных сигналов
WO2011009196A1 (en) * 2009-07-20 2011-01-27 Valorbec, Societe En Commandite Finite dataset interpolation method
US8582869B2 (en) 2009-07-20 2013-11-12 Concordia University Finite dataset interpolation method

Also Published As

Publication number Publication date
CA2263454A1 (en) 1998-12-23
KR20000068164A (ko) 2000-11-25
JPH1173410A (ja) 1999-03-16
EP0932106A1 (en) 1999-07-28
CN1236456A (zh) 1999-11-24

Similar Documents

Publication Publication Date Title
JP4183205B2 (ja) 高速アスペクト比ズーミング及び圧縮及びシーリングを用いるサムネイル操作
JP4002502B2 (ja) 座標インタポレータの符号化/復号化装置及びその方法
US6693645B2 (en) Optimized BIFS encoder
KR20160045613A (ko) 서브밴드 인코딩 된 이미지들로부터 전체 해상도 이미지를 저장, 처리 및 복원하는 방법 및 장치
JP2002500455A (ja) 高速idct/ダウンサンプリング複合演算方法および装置
JP4688988B2 (ja) ビデオデータの圧縮方法並びに装置、及び伸張方法並びに装置
JPH09322165A (ja) 画像復号化装置とその方法、および、画像再生装置
WO1998058328A1 (fr) Procede et dispositif de conversion de signal numerique, dispositif et procede de generation de matrice de transformation et support d&#39;alimentation
JP3701824B2 (ja) データ処理方法、システム、装置、コンピュータ読取り可能媒体及びプログラム記憶装置
JP2001136527A (ja) 直交変換画像の解像度変換装置及び方法
JPH08294119A (ja) 画像符号化/復号化装置
JP3932244B2 (ja) 画像符号/復号方法及びその装置並びにそのプログラムを記録した記録媒体
WO1993021731A1 (fr) Appareil reproducteur d&#39;images
TW440804B (en) Method and apparatus for resizing image information
KR20000029715A (ko) 화상매핑장치및방법과화상생성장치및방법
JP3403724B2 (ja) 画像再生装置及び方法
JP4127465B2 (ja) 画像伸張装置及び方法、並びに記録媒体
JP3198996B2 (ja) 直交変換符号化画像の画像サイズ変換方法
JP2522357B2 (ja) 画像の拡大方式
US7099523B2 (en) Method and system for scaling a signal sample rate
JP2005012460A (ja) 変換符号化係数のサンプリング変換・解像度変換装置
KR100848816B1 (ko) 정수 dct를 이용한 영상 이미지의 크기 변환 방법
JP3709106B2 (ja) 画像圧縮および伸張装置
JP2006340300A (ja) 信号処理方法及び信号処理装置、並びに信号処理プログラム及び情報記録媒体
JP3417761B2 (ja) 画像処理装置

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 98801133.6

Country of ref document: CN

AK Designated states

Kind code of ref document: A1

Designated state(s): CA CN KR US

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE

ENP Entry into the national phase

Ref document number: 2263454

Country of ref document: CA

Ref document number: 2263454

Country of ref document: CA

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 1019997001259

Country of ref document: KR

WWE Wipo information: entry into national phase

Ref document number: 1998924648

Country of ref document: EP

121 Ep: the epo has been informed by wipo that ep was designated in this application
WWE Wipo information: entry into national phase

Ref document number: 09242300

Country of ref document: US

WWP Wipo information: published in national office

Ref document number: 1998924648

Country of ref document: EP

WWP Wipo information: published in national office

Ref document number: 1019997001259

Country of ref document: KR

WWW Wipo information: withdrawn in national office

Ref document number: 1998924648

Country of ref document: EP

WWW Wipo information: withdrawn in national office

Ref document number: 1019997001259

Country of ref document: KR