JPH04229382A - Digital image data resolution exchange device - Google Patents

Abstract

PURPOSE:To convert the resolution of a digital image without adding any special function by utilizing orthogonal conversion for a high-efficiency encoding system for a digital image which has continuous gradations like a full-color image. CONSTITUTION:An original image is divided into mXm picture elements in one block and DCT conversion is performed to obtain conversion coefficients of an mXm matrix. When n<m, components of high spatial frequency among the conversion coefficients of the mXm matrix are deleted horizontally and vertically by >>(m-n) and reverse scale DCT conversion is performed as to the conversion coefficients of the obtained nXn matrix. Consequently, a reproduced image of nXn picture elements in one block is obtained and Xn/m reduction is performed.

Description

[Detailed description of the invention]

0001

[Field of Industrial Application] The present invention relates to a resolution conversion method and apparatus in a high-efficiency encoding system for digital images having continuous gradation such as full-color images, and relates to an image transmission apparatus such as a facsimile, an image storage apparatus such as an image file, etc. It is suitable for application to, etc.

0002

[Prior Art] DCT is used as a high-efficiency encoding method for digital image data, especially continuous tone images such as full-color images.
(Discrete Cosine Transfer
m: discrete cosine transform) is attracting attention. for example,
Video encoding system for video calls/video conferences by CCITT (International Telegraph and Telephone Consultative Committee), CCITT and I
SO (International Organization for Standardization) joint organization JPEG (Join
t Photographic Experts Gr.
oup) color still image encoding method, ISO MP
EG (Moving Picture Experts)
It is planned that a high-efficiency coding method using DCT will be adopted in all of the storage-based video coding methods proposed by the Japanese Yahoo! Group. In addition, there is a growing movement to adopt DCT in digital electronic still cameras, image filing devices using optical disks, and the like.

FIG. 7 is a block diagram showing an example of a color still image encoding device using DCT. In the same figure,
The original image memory 21 is, for example, a memory that encodes an image consisting of 512 x 512 pixels with 8 bits for each pixel and stores it as image data having gradation. For example, it is divided into a plurality of blocks each consisting of 8×8 pixels and is supplied to the DCT circuit 22. In the DCT circuit 22, two-dimensional DC
T is performed to obtain the transform coefficients of the M×N matrix.

DCT is a type of orthogonal transformation in the frequency domain, and converts one block of image data into fij (i=0,1
,...,M-1, j=0,1,...,N-1), and the conversion coefficients are Fuv (u=0,1,...,M-1, v=0,1,...,
N-1), then

[Equation 1] However, Cu , Cv = 1/21/2 (u, v = 0
) =1
It is defined as (u, v≠0). In this case, 8×
Since there are 8 matrices, M=N=8.

[0005] The DCT coefficient Fuv represents a spatial frequency component of one block of image data fij, of which the coefficient F0
0 represents a value (DC component) proportional to the average value of image data fij, and the other coefficients represent alternating current (AC) components, respectively. The coefficients of the AC component represent components with higher spatial frequencies as the variables u and v become larger. DCT coefficients are generally required to have an accuracy of about 11 bits.

8×8 DCs obtained by the DCT circuit 22
The T coefficient is quantized by the quantization circuit 23 by dividing it by each threshold value of a quantization matrix consisting of 8×8 threshold values. Figure 8
shows an example of a quantization matrix for luminance signals.

The quantized DCT coefficients are supplied to the variable length encoding circuit 24, and the difference between the DC component F00 and the DC component quantized in the previous block is taken, and the number of bits of the difference is variable length encoded. be done. The AC component is zigzag scanned within the block from the low frequency component to the high frequency component in the order shown in Figure 9 and converted into a one-dimensional number sequence, and the number of consecutive zero coefficients, that is, the run length, is then Two-dimensional variable length encoding is performed using the number of bits of the subsequent non-zero coefficients (effective coefficients). A Huffman code is used as the variable length code.

The variable length code thus generated is read out and sent to the decoding side through a transmission line or stored in a storage device. On the decoding side, decoding processing is performed in the reverse procedure to that used for encoding. That is, in the variable length decoding circuit 25, the variable length code is decoded in the reverse procedure of encoding to obtain 8×8 quantized DCT coefficients. Then,
In the inverse quantization circuit 26, a quantization matrix similar to that used in quantization is used to obtain 8×8 DCT coefficients Fuv before quantization.

[0009] Then, in the inverse DCT circuit 27, the DCT coefficient Fuv obtained by the inverse quantization circuit 26 is inversely transformed using the transformation formula (2) shown below to obtain image data fij of one block of 8×8 pixels.

[Formula 2] However, Cu, Cv = 1/21/2 (u, v=0
) =1
(u, v≠0). In addition, in this case, M=
N=8.

The image data fij of one block of 8×8 pixels obtained in this way is stored in the reproduced image memory 28,
The original image is restored by combining the original 512×512 pixels in the memory 28.

In the case of a full-color image, assuming that each RGB color has an 8-bit gradation, the above-described operation may be performed for each color. This is YMC, YCb C
The same applies when using other color spaces such as r.

As described above, in the encoding method shown in FIG. 7, the gradation of each pixel in one block of 8×8 pixels is set to 8 bits, and the accuracy of each coefficient after conversion is set to about 11 bits.
0, perform DCT and inverse DCT (IDCT) operations. The image restored by such a series of operations can be restored almost faithfully to the original image.

[0013]

By the way, input/output devices that handle full-color images include television cameras, scanners, display devices, full-color printers, and the like. However, many of these devices differ in the basic functions that make up the image, such as resolution and number of gradations, and in order to transfer image data between these devices, conversion processing is required to absorb differences in resolution, etc. becomes.

One method of resolution conversion processing is to utilize pixel density interpolation. There are several methods for density interpolation, but one of the best methods for interpolating continuous tone images is cubic interpolation.
tion) There is a law. This method uses grid points (x
When finding non-lattice points (uo, vo) from i, yj), the function g(x) = (s
In this method, the concentration of a non-lattice point (uo, vo) is determined from the surrounding lattice points using a three-dimensional polynomial approximation of
), then

[Math. 3] However, D(x) is 3 of g(x) mentioned above.
When interpolating from 16 surrounding grid points using degree polynomial approximation, D(x) = 1-2|x|2
+｜x｜3 (0≦｜x｜<
1) =4-8｜x｜
+5 | x | 2 - | x | 3 (1≦|x|<2)
=0
(2≦
|x|).

However, this method has the disadvantage that it requires a dedicated LSI for real-time processing because there are many surrounding pixels to refer to, which increases the implementation cost such as the amount of calculation. Other interpolation methods include the nearest neighbor method and the linear interpolation method, which are simpler than the cubic interpolation method, but each has the disadvantage that the quality of the reproduced image is inferior.

[0016] Furthermore, in an apparatus for transmitting and storing full-color images using a high-efficiency image encoding system, the above-mentioned DC
Since it is necessary to add a resolution conversion function in addition to the image encoding function using T, the system cost increases, which is inconvenient.

The present invention provides a method and apparatus for resolution conversion of digital image data, which has a resolution conversion function without adding any special functions, in a high-efficiency encoding system for digital images having continuous gradations such as full-color images. The purpose is to provide

[0018]

[Means for Solving the Problems] A resolution conversion method for digital image data according to the present invention performs orthogonal transformation on a two-dimensional pixel block of a first size, and converts a block of two-dimensional coefficients of the first size into a first size. The present invention is characterized in that a block of two-dimensional coefficients of a size of 2 is generated, and a two-dimensional pixel block of a second size is obtained by performing an inverse transformation of the orthogonal transformation on this block.

The resolution conversion device for digital image data according to the present invention also includes encoding means for compressing and encoding an original image represented by digital image data and transmitting or storing the compressed image data, and decompressing and decoding the compressed and encoded image data. and a decoding means for obtaining a reproduced image using a block of m×n pixels.
means for obtaining p x q transform coefficients from the m x n transform coefficients, quantizing the p x q transform coefficients, and varying the quantized transform coefficients; The decoding means performs variable length decoding on the image data compressed and encoded by the encoding means, and dequantizes the obtained p×q transform coefficients; means for performing two-dimensional orthogonal transformation on the p x q transform coefficients obtained by the inverse quantization to obtain image data of one block of p x q pixels; and a means for generating a reproduced image by combining the image data of the p x q pixel block. It is characterized by consisting of a means for obtaining.

[0020]

[Operation] The resolution conversion method and device according to the present invention have the following features:
First, digital image data of an original image is divided into blocks of a first size, for example, m×m pixels, and orthogonal transformation is performed for each block to obtain m×m transform coefficients. Then,
To obtain a second size, e.g., an n×n pixel block, if n<m, reduce the high spatial frequency components of the m×m transform coefficients by “m−n” in the horizontal and vertical directions. Then, inverse transform is performed on the obtained n×n transform coefficients. As a result, a reproduced image of n×n pixel blocks reduced by n/m times is obtained.

[0021] If m<n, the transform coefficients of the m x m matrix are extrapolated to the side with higher spatial frequency components by "n-m" coefficients in the horizontal and vertical directions. Inverse transform is performed on the transform coefficients of the n×n matrix. As a result, a reproduced image of n×n pixel blocks enlarged by n/m times is obtained.

Further, if the first size is an m×n pixel block and the second size is a p×q pixel block, different magnifications such as p/m times in the horizontal direction and q/n times in the vertical direction are obtained. resolution conversion is possible. In this case, for example, conversion such as expansion in the horizontal direction and reduction in the vertical direction is also possible.

Therefore, according to the present invention, since resolution conversion can be performed using orthogonal transformation, a high-efficiency image encoding device using orthogonal transformation can perform resolution conversion without adding any special functions. can be done.

[0024]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A method of converting the resolution of digital image data according to the present invention will be described with reference to FIGS. 1 to 5. In the color still image encoding method using DCT described above, as shown in FIG. 10, transform coefficients F00 to F77 of one block of 8x8 matrix are converted to image data of 8x8 pixels of one block f00 to f00 by inverse DCT. For example, as shown in FIG. 1, among the conversion coefficients F00 to F77 of the 8×8 matrix,
Using only the conversion coefficients F00 to F33 of 4 matrices,
If M=N=4 in equation (2) during inverse DCT, a reproduced image can be obtained from image data f00 to f33 of one block of 4×4 pixels.

In principle, this corresponds to resampling an 8×8 pixel block image into a 4×4 pixel block image by reducing the sample rate to 1/2. In addition, in this method, since coefficients of components having high spatial frequencies other than the transform coefficients F00 to F33 are deleted, aliasing distortion that normally occurs due to resampling can be prevented, and a filtering effect can be achieved at the same time. This process reduces or converts the resolution of a 512 x 512 pixel image to a 256 x 256 pixel image.

Furthermore, in similar processing, for example, FIG.
As shown in Figure 2, a coefficient "0" is added around the 8x8 matrix transform coefficient (extrapolation) to generate a 12x12 matrix transform coefficient, and inverse transformation is performed as M=N=12 in equation (■). By doing so, it is possible to obtain a reproduced image of 1 block of 12×12 pixels. With this process, 51
A 2 x 512 pixel image is enlarged or resolution converted to a 768 x 768 pixel image. Note that the coefficient for extrapolation is “
In addition to 0'', other extrapolation methods such as linear prediction may be used.

When these processes are generalized, the above-mentioned DC
In the T-transform equation (■), set M=N=m, and perform the above-mentioned inverse DCT
If M=N=n in the conversion equation (2), it shows that resolution conversion of n/m times is possible.

To explain in more detail with reference to FIGS. 3 and 4, first, in FIG. 3, the original image is divided into one block m
Divide into xm pixels (a in the same figure), and in the formula (■), M=N=m
DCT transform is performed as follows to obtain transform coefficients of an m×m matrix. Next, in the case of n<m, among the conversion coefficients of the m×m matrix, components with high spatial frequencies are reduced by “m−n” in the horizontal and vertical directions (see figure b), and the resulting n×n
Regarding the conversion coefficient of the matrix, in the formula ■, M=N=n
Inverse DCT transformation is performed as follows. As a result, 1 block n×
A reproduced image of n pixels is obtained (c in the same figure) and can be reduced by n/m times.

In addition, in the case of m<n, in FIG.
About the conversion coefficients of the n×n matrix obtained by extrapolating "n-m" coefficients horizontally and vertically to the higher spatial frequency component among the transform coefficients of the ×m matrix (see figure b). In the equation (2), inverse DCT transformation is performed with M=N=n. As a result, a reproduced image of 1 block of n×n pixels is obtained (c in the same figure), and the image can be enlarged by a factor of n/m.

Furthermore, as shown in FIG. 5, in the DCT transformation formula (■), if M=m, N=n, and in the inverse DCT transformation formula (■), M=p, N=q, then p/m in the horizontal direction. It is possible to perform resolution conversion at different magnifications, such as q/n times in the vertical direction. In this case, for example, conversion may be performed such as expansion in the horizontal direction and reduction in the vertical direction. This method is effective in, for example, resolution conversion between encoding devices with different aspect ratios.

In the above explanation, some of the transform coefficients F00 to F77 are deleted or extrapolated during decoding,
I tried converting the resolution, but when encoding, the conversion coefficient F
The resolution may be converted by partially deleting 00 to F77 or performing extrapolation.

Next, an embodiment in which the resolution conversion device for digital image data according to the present invention is applied to a high-efficiency encoding device for still images will be described with reference to FIG. This embodiment shows an example in which part of the transform coefficients are deleted or extrapolated during encoding.

In FIG. 6, if the color scanner 1 has a resolution of, for example, A4 size 200 dpi, it can read an image of approximately 2376×1728 pixels on one screen. On the other hand, if the color CRT 2 has a resolution of 768 x 488 pixels per screen, for example, the resolution ratio is 768/2376 in the horizontal direction,
The vertical direction is 488/1728, and when expressed in simple integer ratios, the horizontal direction is 3/9 (=p/m) and the vertical direction is 3/11 (
=q/n) is appropriate. Therefore, in this case, m×n is 9×11 in the conversion method shown in FIG.
, p×q can be set as 3×3.

In FIG. 6, image data read from the color scanner 1 is temporarily written into the first buffer memory 3. The data on the buffer memory 3 is read out for each m×n pixel block by the blocking circuit 4 and supplied to the DCT circuit 5, where it is subjected to two-dimensional DCT for each block as described above, and is converted into an m×n matrix. Get the conversion coefficients.

Next, the matrix conversion circuit 6 converts m
Processing is performed to convert the transformation coefficients of the ×n matrix into the transformation coefficients of the p×q matrix. In this example, since one block is converted from 9×11 (m×n) to 3×3 (p×q), it is reduced in both the horizontal and vertical directions, and the procedure in FIG. 3 described above is applied. Further, at this time, the values of m, n, p, and q in the blocking circuit 4, DCT circuit 5, and matrix conversion circuit 6 are set by the blocking control circuit 7.

The thus obtained p×q matrix transform coefficients are then quantized by a quantization circuit 8, further Huffman encoded by a variable length encoding circuit 9, and transmitted or stored.

Decoding is performed by Huffman decoding in the variable length decoding circuit 10, inverse quantization in the inverse quantization circuit 11, and inverse DCT of the transform coefficients of the p×q matrix in the inverse DCT circuit 12. Obtain pixel image data. This data is temporarily stored in the second buffer memory 13.
By writing the image data into the CRT 2 and reading it when all the blocks of image data have been written, the image data read by the color scanner 1 can be displayed at the resolution of the CRT 2.

[0038] Also in decoding, each value of p and q is controlled by the blocking control circuit 14. Also, m,
Each value of n, p, and q depends on the resolution of the input device on the encoding side (color scanner 1) and the output device on the decoding side (color CRT).
2) is calculated as described above by a control section (not shown) and stored in the blocking control circuits 7 and 14 in advance.

In the above-described embodiments, the present invention was applied to a still image high-efficiency encoding method and apparatus, but the essential part of the present invention is related to orthogonal transformation processing. It is clear that this can also be applied to video encoding systems.

[0040]

[Effects of the Invention] According to the present invention, resolution conversion using orthogonal transformation becomes possible, so it can be immediately applied to a high-efficiency image coding method that uses orthogonal transformation without adding any special functions. This makes it possible to transfer image data between devices with different resolutions and numbers of gradations without increasing system costs.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an example of 8×8 pixel/4×4 pixel conversion according to the present invention.

FIG. 2 is a diagram showing an example of 8×8 pixel/12×12 pixel conversion according to the present invention.

FIG. 3: m×m pixels/n×n pixels (n<
m) A diagram showing an example of conversion.

FIG. 4: m×m pixels/n×n pixels (m<
n) is a diagram showing a conversion example.

FIG. 5 is a diagram showing an example of m×n pixel/p×q pixel conversion according to the present invention.

FIG. 6 is a block diagram of a high-efficiency still image encoding device to which the present invention is applied.

FIG. 7 is a block diagram of a conventional color still image encoding device using DCT.

FIG. 8 is a diagram showing a quantization matrix.

FIG. 9 is a diagram showing a zigzag scan table.

FIG. 10: 8×8 pixels/8×8 in the device shown in FIG. 7
FIG. 3 is a diagram showing an example of pixel conversion.

FIG. 11 is an explanatory diagram of a density interpolation method for non-lattice points.

[Explanation of symbols]

1 Color scanner 2 Color CRT 3, 13 Buffer memory 4 Blocking circuit 5 DCT circuit 6 Matrix conversion circuit 7, 14 Blocking control circuit 8 Quantization circuit 9 Variable length encoding circuit 10 Variable length decoding circuit 11 Dequantization circuit 12 Inverse DCT circuit

Claims (8)

[Claims] Claim 1: Perform orthogonal transformation on a two-dimensional pixel block of a first size, generate a block of two-dimensional coefficients of a second size from the resulting block of two-dimensional coefficients of the first size, and generate a block of two-dimensional coefficients of a second size; A method for converting the resolution of digital image data, characterized in that a two-dimensional pixel block of the second size is obtained by performing an inverse transformation of the orthogonal transformation. 2. In claim 1, the generation of the second-sized two-dimensional coefficient block is performed by generating high spatial frequency components from the first-sized two-dimensional coefficient block after the orthogonal transformation. A method for converting the resolution of digital image data, characterized by removing side coefficients and generating only the remaining coefficients. 3. In claim 1, the generation of the two-dimensional coefficient block of the second size is performed by generating high spatial frequency components from the two-dimensional coefficient block of the first size after the orthogonal transformation. A method for converting the resolution of digital image data, characterized in that the resolution conversion method of digital image data is generated by interpolating coefficients on the side. 4. When an original image represented by digital image data is divided into blocks of m×m pixels, subjected to orthogonal transformation, and then reduced to n×n pixels (n<m), the spatial frequency is A digital image characterized by performing n/m-fold resolution conversion by deleting "m-n" coefficients on the higher component side in the horizontal and vertical directions and then performing n×n inverse transformation. Data resolution conversion method. 5. When an original image represented by digital image data is divided into blocks of m×m pixels, subjected to orthogonal transformation, and then enlarged to n×n pixels (m<n), the spatial frequency is "m" horizontally and vertically on the high component side, respectively.
1. A method for converting the resolution of digital image data, characterized in that the resolution is converted by a factor of n/m by interpolating the coefficients by "-n" coefficients and then performing an inverse conversion of n×n. 6. Block an original image represented by digital image data into m×n pixels, perform orthogonal transformation on this, generate a p×q coefficient block from the resulting m×n coefficient block, and A method for converting the resolution of digital image data, the method comprising performing p×q inverse transformation to obtain p×q pixel blocks. 7. An encoding device for compressing and encoding an original image represented by digital image data and transmitting or storing the compressed image data; and a decoding device for decompressing and decoding the compressed and encoded image data to obtain a reproduced image, The encoding means includes means for dividing the original image into one block of m×n pixels, means for performing two-dimensional orthogonal transformation on the m×n pixel block to obtain m×n transform coefficients, and It consists of means for obtaining p×q transform coefficients from ×n transform coefficients, and means for quantizing the p×q transform coefficients and subjecting the quantized transform coefficients to variable length encoding, The decoding means performs variable length decoding on the image data compressed and encoded by the encoding means, and dequantizes the obtained p×q transform coefficients, and the p×q transform coefficients obtained by the dequantization. The method comprises means for performing two-dimensional orthogonal transformation on q transform coefficients to obtain image data of one block of p×q pixels, and means for obtaining a reproduced image by combining the image data of the p×q pixel blocks. A resolution conversion device for digital image data. 8. A resolution conversion method and apparatus for digital image data according to claim 1, wherein the orthogonal transformation is a discrete cosine transformation.