JP2003102012A - Image encoding device, image decoding device, image encoding method, and image decoding method - Google Patents

Abstract

(57) Abstract: An image encoding method and apparatus, and an image decoding method and apparatus capable of achieving a higher compression ratio than when directly outputting binary symbols are obtained. In particular, based on information as to which of the binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol, a dominant symbol is set from a Huffman code set systematically created for an extended information source of the binary symbol. And a binary symbol encoding unit 704 that selects a code optimal for the situation of the extended information source of the binary symbol assumed from the estimated occurrence probability of the binary symbol and performs binary information source encoding, and which of the binary symbols is the dominant symbol From the Huffman code set systematically created for the binary symbol extended information source based on the information and the estimated appearance probability of the superior symbol, the extended information of the binary symbol assumed from the estimated appearance probability of the superior symbol. A binary symbol decoding unit 801 for selecting a code most suitable for the situation of the source and performing binary information source decoding.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image encoding method and apparatus for encoding an image signal, and an image decoding method and apparatus for decoding an encoded signal.

[0002]

2. Description of the Related Art As an image coding method using wavelet transform, for example, "A new, fast, and
efficient image coding b
based on set partitioning
in hierarchical trees "
(A. Said, WA Pearlman, IEEE
transactions on circuits
and systems for video tec
hology, Vol. 6, No. 3, 1996). In this method, the wavelet transform coefficient is represented by a bit plane composed of absolute values and a sign bit representing positive / negative, and encoded in order from the upper bit plane.
At this time, So as the first pass for encoding the coefficient (insignificant coefficient) in which 1 has not yet occurred in the upper plane
The processing is divided into two types, namely, Passing Pass and Refinement Pass as the second pass for encoding the coefficient (significant coefficient) in which 1 has already occurred in the upper plane.

In FIG. 9, an image is subjected to three-stage wavelet transform, and then the LL component (horizontal low-pass filter, vertical low-pass filter) of the transform coefficient is arranged at the upper left, and thereafter, each resolution is set. HL component (horizontal high-pass filter, vertical low-pass filter), LH component (horizontal low-pass filter, vertical high-pass filter), HH component (horizontal high-pass filter, vertical high-pass filter) It is a figure arranged in order from the resolution.

Here, in Sorting pass, attention is paid to the tree structure of coefficients corresponding to the same frequency component at the same spatial position (eg, a tree of coefficients hatched in FIG. 9), and the tree vertices are defined as the tree vertices. All the conversion coefficients equal to or lower than a certain conversion coefficient are output as binary symbols 0/1 indicating whether or not they are insignificant coefficients. If all are non-significant coefficients, the coding of the transform coefficients belonging to that tree in that bit plane is complete. However, if all are not non-significant coefficients, the tree of transform coefficients is divided into four, that is, four trees whose vertexes are the four coefficients one level below the current vertex of the tree. Next, it is checked again whether or not all the divided trees are non-significant coefficients.
Repeat this procedure for all non-significant coefficients.

At this time, a list in which the coordinates of the transform coefficients are registered is prepared, and the transform coefficients are encoded in the order according to the list. There are the following three types of lists. LSP (List of Significant P
(ixel): a list LIP (List of SIGNIFICANT) for registering the coordinates of the significant coefficient
Pixel): List LIS (List of Significant S) for registering coordinates of insignificant coefficients
ets): a list that registers the coordinates of the vertices of the tree that includes the significant coefficient

In each bit plane, Sortin
g Pass first encodes the significance / insignificance of the coefficients registered in LIP and then examines the significance / insignificance of the entire tree registered in LIS. If the tree contains significant coefficients, the tree is split and the LIS is updated to see if the split tree is significant / insignificant.

As for the refinement pass, the registered coefficient (however, So of the same plane is used)
Excluding the coefficient that became a significant coefficient in the rending pass)
The bit in the relevant bit plane of is output as a code.

10 to 12 are flowcharts showing the procedure of the encoding process. Here, the absolute value of the conversion coefficient at the coordinates (i, j) is C (i, j), and the conversion is one level lower corresponding to the frequency component in the same spatial position and in the same direction as C (i, j). Coefficients are T (i, j), C (i, j)
The set of all lower-level coefficients corresponding to frequency components spatially at the same position and in the same direction as D (i, j), C
A set of all the coefficients of two levels or lower, which correspond to frequency components spatially at the same position and in the same direction as (i, j), are represented by L (i, j). Also,

[0009]

[Equation 1]

[0010] Here, n is n = log ₂
max C (i, j), and becomes smaller by 1 as the bit plane becomes lower.

In this method, the binary symbol generated in the above processing is output as it is as a code. Further, since the information is transmitted in order from the important information of the image information, that is, from the upper plane and from the low resolution component in the plane, even if the encoding is finished at any point in the encoding process, it is generated. The coded data provides an almost optimum image quality obtained with the coded amount. The decryption process is shown in FIGS.
This can be realized by replacing "output" in FIG. 12 with "input" for processing.

[0012]

By the way, in the image coding method shown in the above-mentioned conventional example, the generated binary symbol is output as a code as it is, so that the compression performance is insufficient in some cases. There was a point.

The present invention has been made in order to solve the above problems, and an image coding apparatus and an image which can achieve a higher compression ratio than a case where a generated binary symbol is directly output. An object is to provide a decoding device, an image encoding method, and an image decoding method.

[0014]

An image coding apparatus according to the present invention comprises a wavelet transform means for wavelet transforming an image signal, a transform coefficient from the wavelet transform means, a bit plane composed of absolute values thereof, and a positive / negative sign. In each of the above bit planes, a binary symbol representing a bit in a transform coefficient equal to or less than a threshold value, which is the first pass, is output in the bit plane forming unit that represents A binary symbol output means for outputting a binary symbol representing a bit in a transform coefficient, and information on which of the above binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol are used as an expanded information source of the binary symbol. It is assumed from the estimated appearance probability of the dominant symbol from the Huffman code set created systematically. It is obtained by a binary symbol encoding means for performing by selecting an optimal code in the context of expansion source value symbol binary source coding.

Further, the binary symbol encoding means converts the binary symbol output from the binary symbol output means into frequency components in the same spatial position and in the same direction in the first pass. A binary symbol sequence that represents whether or not all corresponding sets of conversion coefficients of a certain resolution or higher are below a certain threshold value, or 2 that represents whether or not a certain conversion coefficient is below a certain threshold value
A binary symbol sequence that represents the positive / negative of the conversion coefficient is classified into binary symbol sequences, and each binary sequence is independently binary symbol-encoded according to the context, and in the second pass, the binary symbol output is performed. The binary symbol output from the means is independently binary-source coded as one binary symbol sequence.

In the binary symbol coding means, the codeword in a certain binary symbol sequence is fixed in the classified plurality of binary symbol sequences, and the codeword in another binary symbol sequence is determined. Is determined, the output order of the codeword is changed.

Further, the binary symbol coding means virtually inserts a dummy binary symbol even if the code word is not fixed in the last binary symbol of each path in the bit plane. The code is fixed.

Further, the binary symbol encoding means deletes all the bits exceeding the set code amount from the generated code when the generated code amount exceeds the set code amount.

Further, the image decoding apparatus according to the present invention determines the binary symbol expanded information source based on the information indicating which of the coded binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol. On the other hand, from the Huffman code set created systematically, a binary symbol for decoding the binary information source by selecting the optimal code for the situation of the extended information source of the binary symbol assumed from the estimated appearance probability of the dominant symbol A decoding means, a binary symbol of a first pass representing bits in a transform coefficient equal to or less than a certain threshold value, and a binary symbol of a second pass representing bits in a transform coefficient other than that, and an absolute value of the transform coefficient. Bit plane reproducing means for reproducing the sign bit representing the positive and negative and the bit plane constituted by the absolute value of the above conversion coefficient and the positive and negative signs. A conversion coefficient reproducing means for reconstructing the transform coefficients from the sign bit, in which a wavelet inverse transform means for inverse wavelet transform the transform coefficients.

In addition, in the first pass, the binary symbol decoding means determines whether all sets of transform coefficients having a certain resolution or more corresponding to frequency components in the same spatial position and in the same direction are less than a threshold value. A binary symbol sequence indicating whether or not a certain conversion coefficient is less than a certain threshold value, a binary symbol sequence indicating whether the conversion coefficient is positive or negative, and a binary symbol sequence indicating whether the conversion coefficient is positive or negative, and each binary sequence is classified into its context. Accordingly, binary symbol decoding is independently performed, and in the second pass, one binary symbol sequence is independently binary information source decoded.

Further, the above-mentioned binary symbol decoding means, even if the decoded binary symbol remains without being used for reproduction of the transform coefficient at the end of each pass in the bit plane,
The dummy binary symbol is determined and the dummy binary symbol is discarded.

Further, the binary symbol decoding means is a code word formed by adding bits corresponding to the number of insufficient bits to the short code word when the code word located at the end of the code data is shorter than the regular code word. Only the binary symbol part that is common to all the candidates of the decoded binary symbol sequence corresponding to is the decoded binary symbol.

The image coding method according to the present invention is
Wavelet transforming the image signal, expressing the transform coefficient obtained by the wavelet transform with a bit plane composed of absolute values and a sign bit representing positive or negative; in each of the bit planes, in the first pass The step of outputting a binary symbol that represents a bit in a conversion coefficient equal to or less than a certain threshold value and outputting a binary symbol that represents a bit in another conversion coefficient in the second pass, and which of the binary symbols is the dominant symbol? Extended information source of binary symbol assumed from estimated appearance probability of dominant symbol from Huffman code set systematically created for extended information source of binary symbol based on information and estimated occurrence probability of dominant symbol And the step of performing binary information source coding by selecting the most suitable code for the situation That.

Further, in the step of performing the binary information source coding, the binary symbol output in each bit plane has a resolution higher than a certain resolution corresponding to the same spatial position in the first pass. A set of conversion coefficients is classified into a binary symbol series that indicates whether or not a conversion coefficient is less than a certain threshold, a binary symbol series that indicates whether or not a certain conversion coefficient is less than a certain threshold, and a binary symbol series that indicates whether the conversion coefficient is positive or negative. , Each 2
The value sequence is independently binary-coded according to the context, and in the second pass, the binary symbols output in each bit plane are independently binary-coded as one binary symbol sequence. To do.

Further, in the step of performing the binary information source coding, in a plurality of classified binary symbol sequences, a code word in a certain binary symbol sequence is fixed, and another binary symbol sequence is determined.
When the codeword in the value symbol sequence is not fixed, the output order of the codeword is changed.

In the binary information source coding step, a dummy binary symbol is virtually inserted even if the code word is not fixed in the last binary symbol of each path in the bit plane. As a result, the code is fixed.

The step of performing the binary information source coding is to delete all the bits exceeding the set code amount from the generated code when the generated code amount exceeds the set code amount.

Further, the image decoding method according to the present invention determines the binary symbol expansion information source based on the information indicating which of the coded binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol. On the other hand, from the Huffman code set created systematically, a step of selecting a code most suitable for the situation of the extended information source of the binary symbol assumed from the estimated appearance probability of the dominant symbol and performing the binary information source decoding, A bit plane composed of an absolute value of a transform coefficient from a binary symbol of a first pass that represents a bit of a transform coefficient equal to or less than a threshold and a binary symbol of a second path that represents a bit of another transform coefficient. And a step of reproducing a sign bit representing positive and negative, and a bit plane composed of the absolute value of the above transform coefficient and a transform coefficient from the sign bit representing positive and negative A step of reconstructing, in which a step of inverse wavelet transform the transform coefficients.

Further, the step of performing the binary information source decoding indicates whether or not all sets of transform coefficients having a certain resolution or higher corresponding to the same spatial position are less than a certain threshold in the first pass. A binary symbol sequence, a binary symbol sequence that indicates whether or not a certain conversion coefficient is less than a certain threshold, and a binary symbol sequence that indicates whether the conversion coefficient is positive or negative,
Binary information source decoding is independently performed on the value sequence according to the context, and in the second pass, one binary symbol sequence is independently binary information source decoding.

In the step of performing the binary information source decoding, even if the decoded binary symbol remains without being used for reproducing the transform coefficient at the end of each pass in the bit plane, a dummy binary symbol is used. 2 of the dummy
It discards the value symbol.

In the step of performing the binary information source decoding, when the code word located at the end of the code data is shorter than the regular code word, the short code word is added with bits corresponding to the number of insufficient bits. Only the binary symbol part common to all the candidates of the decoded binary symbol sequence corresponding to the possible codeword is used as the decoded binary symbol.

[0032]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below.
It will be described with reference to the drawings. Embodiment 1. 1 is a block diagram showing a configuration of an encoding method and apparatus according to Embodiment 1 of the present invention.
In FIG. 1, reference numeral 701 denotes a wavelet transform unit as a wavelet transform means for performing wavelet transform, 7
Reference numeral 02 denotes a bit plane conversion unit as a bit plane conversion unit that expresses the wavelet transform coefficient with a bit plane composed of its absolute value and with a sign bit that indicates whether the sign is positive or negative.
Reference numeral 3 is a bit plane, a binary symbol output section as a binary symbol output means for converting sign bits into binary symbols, and 704 is a binary symbol encoding means as a binary symbol encoding means for entropy encoding the binary symbols. It is a department.

FIG. 2 is a block diagram showing the configuration of the decoding method and device according to the first embodiment of the present invention. Figure 2
801 is a binary symbol decoding unit as a binary symbol decoding means for entropy decoding a binary symbol, and 8
Reference numeral 02 denotes a bit plane of a transform coefficient from a binary symbol, a bit plane reproducing unit as a bit plane reproducing means for generating a sign bit, 803 denotes a bit plane of a transform coefficient, and a transform coefficient reproducing means for reproducing a transform coefficient from a sign bit. A transform coefficient reproducing unit 804 is a wavelet inverse transforming unit as a wavelet inverse transforming unit that performs inverse wavelet transform on the transform coefficient.

Next, the operation will be described with reference to FIGS. 3 and 4.

First, on the encoding side shown in FIG. 1, a wavelet transform unit 701 performs a wavelet transform on an image signal, and a bit plane conversion unit 702 outputs a wavelet transform coefficient to a bit plane composed of its absolute values and a sign representing positive and negative. Express in bits. Then, the binary symbol output unit 703 converts the bit plane and the sign bit into binary symbols and outputs them, and the binary symbol encoding unit 704 entropy-encodes the binary symbols.

Here, in the binary symbol encoding unit 704,
Is one or more binary symbols, as shown in FIG.
Assign one codeword to the column. That is, encoding
Is a certain number (hereinafter, this certain number is called the code order)
Of consecutive binary symbols "0" of, or
When the binary symbol "1" appears, the code word is fixed and
The code word of is output. However, in this example, the binary symbol
MPS (= More Probeable Sy)
mbol, superior symbol) Binary symbol "1" LPS
(= Less Probable Symbol, recessive
Symbol). At this time, the number of consecutive MPS is
Counted by a counter inside (or outside) the encoder
It The code order can be any natural number
But in this example 2 ⁿThe description will be limited to (2 to the nth power).
The number of consecutive MPS appearances (MPS counter value) is the code order.
If they are equal, assign a 1-bit codeword
It

When an LPS has appeared by then, the number of consecutive appearances of MPS from the last output of the codeword to the appearance of the LPS is M before the n-bit binary value.
Added 1 bit to distinguish it from PS-only codeword (n +
1) A code word of bits is assigned. The unit of the binary symbol sequence to which the code word is assigned is called a message, and the MPS counter is reset at the same time when the code word is determined and output. In this way, a code is obtained by outputting each output codeword as a series of sequences.

On the other hand, on the decoding side shown in FIG. 2, the binary symbol decoding section 801 entropy-decodes the binary symbols. This decoding is performed by decomposing an input code sequence into codewords and restoring a binary symbol sequence for each decoder. Next, a bit plane reproduction unit 802 generates a conversion coefficient bit plane and a sine bit from the binary symbol, a conversion coefficient reproduction unit 803 reproduces a conversion coefficient bit plane and a sine bit, and a wavelet inverse conversion unit. In 804, the transform coefficient is subjected to inverse wavelet transform to reproduce the image.

In the above code, it is possible to realize a further excellent coding efficiency by switching the code order to an appropriate value according to the probability of occurrence of 0/1 estimated from the past binary symbol sequence. it can. An example of the state transition method for determining the code order is shown below. The encoder / decoder is in one of the 32 states shown in FIG. 4 in the process of encoding / decoding a binary symbol sequence, and the code order is determined according to the state. When the code word is fixed, a state transition is executed, and if the number of consecutive MPS occurrences becomes equal to the code order, the state number is incremented by one, and if the LPS has appeared by then, the state number is incremented by one. Lower. However, when the number of consecutive appearances of MPS in the state number 31 becomes equal to the code order, or when the LPS appears in the state number 0, the state is left as it is without transition. Here, the initial value of the state number in the encoder / decoder is 0. Also, at the start of the encoding / decoding process, the MPS counter of each encoder / decoder is reset.

Further, as another example of the method of determining the code order, there is a method of counting the numbers N (0) and N (1) of 0s and 1s on the transmission / reception side and performing it based on the counting result. As the calculation method, 2 ^{n + 1} N (1)> N (0) ≧ 2 ⁿ N
According to (1). However, the code order 2 ^{n that} is the state transition destination
Is less than or equal to the default maximum value and greater than or equal to the default minimum value.

As described above, in the present embodiment, the binary symbol is directly output by using the binary symbol coding method in which the binary symbol as described above is adapted according to the appearance probability. A higher compression rate can be achieved than in the case.

Embodiment 2. The present embodiment is a case where a binary symbol calculated in the same manner as described above is divided into a plurality of binary sequences according to the context, and binary entropy encoding / decoding is independently performed.

5 to 7 are diagrams showing examples of binary symbol output in the second embodiment of the present invention. The basic configurations of the encoder and the decoder are the same as those in the above-described first embodiment, and thus the description thereof is omitted.

Next, the processing procedure of this embodiment will be described with reference to FIGS.
Description will be made using the binary symbol output example shown in FIG. FIG. 5 shows insignificance / significance in the most significant bit plane of the transform coefficient after performing the two-step wavelet transform on the image having the size of 8 × 8 pixels. The LL component (horizontal low-pass filter, vertical low-pass filter) is arranged in the upper left, and thereafter, the HL component (horizontal high-pass filter, vertical low-pass filter), LH component (horizontal low-pass filter) of each resolution , Vertical high-pass filter) and HH components (horizontal high-pass filter, vertical high-pass filter) are arranged in order from low resolution, and the position of each transform coefficient is expressed using coordinates (i, j). I shall. In FIG.
1 indicates a coefficient that is significant in this bit plane, and 0 indicates a coefficient that is not significant in this bit plane, and the sign of the significant coefficient is represented by +/-.

FIG. 6 and FIG. 7 show an output example of binary symbols in Sorting Pass in this bit plane. In FIGS. 6 and 7, the left three columns show the coordinates of the conversion coefficients registered in the LSP, LIP, and LIS. The coefficients erased by the strikethrough are the coefficients deleted once registered. Represent Also, 2 from the right end
The 6th columns from the 7th column to the 7th column show binary symbols output in each context. From left to right: 1) Significance / insignificance of coefficients registered in LIP, 2) Sign when coefficient is significant. 3) Significance / non-significance of coefficient tree belonging to Type-A registered in LIS, 4) Significance / non-significance of lower level 4 coefficients when coefficient tree is significant, 5) Sign when coefficient is significant, 6) Represents a binary symbol indicating significant / insignificant of the coefficient tree belonging to Type-B registered in LIS.

In the present embodiment, these six are set as independent contexts, the degree is set for each of them, and encoding is performed. Hereinafter, these six contexts will be referred to as context 1 to context 6 in order. However,
In order to simplify the explanation, learning of the orders is not performed, and the orders in the contexts 1 to 6 are 4, 1, 2, 4, 1, respectively.
2, MPS has a fixed value of 0, and LPS has a fixed value of 1. At the time of actual coding, in order to improve the coding efficiency, learning of the order shown in the first embodiment is performed.

FIG. 6 and FIG. 7 show examples in which the binary symbols generated in each context are coded using the binary entropy coding method shown in the first embodiment.
For example, the codeword "101" is output for the binary symbol in the context 1 in the processing steps 1 and 2 and the codeword "1" is output for the binary symbol in the context 2 in the processing step 3 in the figure. To be done. The present embodiment is characterized in that the following processing is performed when encoding is continued in such a procedure.

Codeword transmission order control in the coding start order In order to explain the codeword transmission order control, FIG. 6 and FIG.
The processing steps 29 to 37 of FIG. In processing step 29, 0 is generated in context 3,
In processing step 29, the codeword is not fixed. In the processing from the processing step 30 onward, the codeword is determined in the context 4 or 5, but the decoding side must acquire the codeword generated in the context 3 before the codeword generated in the contexts 4 and 5. If it cannot be correctly decoded, the encoding side performs transmission order control to output the code words in the order of occurrence of the first encoded binary symbol.

In the transmission order control, when the code word in a certain context is determined, it is necessary to detect whether or not there is a context in which the leading coded binary symbol is generated first but the code word is not yet determined, and it must be detected. For example, the determined code word is output as it is. If there is a context in which the codeword is not fixed, the codeword is temporarily stored in the code buffer, and the head coded binary symbol is output after the codeword of the context in which the first coded binary symbol occurs is fixed. .

By this processing, the information source is divided into a plurality of contexts, the coding suitable for each context becomes possible, and a higher compression rate can be achieved.

Explaining using the example of FIG. 8, although the binary symbol "0" is generated in the processing step 29, the M of the encoder is used.
Only the PS counter is incremented, and the codeword cannot be determined. Next, in the processing steps 30 to 36, the code words of the contexts 4 and 5 are successively determined, but the code words in the context 3 in which the first encoded binary symbol occurs first have not been finalized, so the code words are not determined. Since it cannot be output, the code words of processing steps 30 to 36 are temporarily stored in the code buffer. Then, after the code word of context 3 is determined in processing step 37, the temporarily stored code words of processing steps 30 to 36 are output.
That is, the codewords are output in the order of to shown in FIG.

On the decoding side, no special processing regarding the transmission order of the code words is required, and a memory for storing the decoded binary symbols is prepared for the number of contexts, and binary data for reproducing the transform coefficient is used in each context. When a symbol is needed, the binary symbol is read from the memory of the context. At that time, if there is no binary symbol to be read in the memory, the next code word is read, the decoded binary symbol is stored in the memory of the context, and the necessary binary symbol is read from the memory to reproduce the transform coefficient. to continue.

Code determination at the end of the pass Sorting at the end of FIG. 7, that is, processing step 69
At the end of Pass, the binary symbol 0 is generated in context 4, but the codeword cannot be determined only by this symbol. In this case, in context 4, a dummy binary symbol "0" is inserted after this, and the code word "0" is output. Here, if the code word cannot be determined in the processing step 69, the Re side that follows this at the decoding side
If the codeword generated in context 4 is not acquired before the codeword generated in the fineness pass, the correct decoding cannot be performed.
The codeword in tpass must be temporarily stored in the memory and output after the codeword in context 4 is fixed.

In this way, by forcibly determining the code at the end of the pass, it is possible to reduce the memory capacity required for temporary storage. Also, if the learning of the order of binary symbol coding is not performed, or if the learning of the order is ended at the end of the pass and the order is set to a predetermined initial value at the beginning of the next pass, The coding result of the subsequent refinement pass does not affect the context generation of the sorting pass of the next plane and the learning of the degree.

Therefore, if the sign is determined at the end of the pass,
In consideration of parallel processing using hardware, since the coding of the sorting pass of the next plane can be started immediately after the sorting pass of the current plane is completed, the processing speed can be increased.

In this embodiment, processing step 5 in FIG.
Similarly, in, the code word is fixed by inserting a dummy binary symbol. This is because the codes of contexts 1 and 2 are generated only at the beginning of the Sorting Pass and do not occur thereafter, so that the code word is fixed in processing step 5 and the trouble of temporarily storing the subsequent code words is saved.

At the time of decoding, the dummy binary symbol inserted at the time of encoding is additionally decoded.
After using the binary symbols necessary for decoding, the correct decoding can be performed by discarding the remaining binary symbols.

6 and 7 show Sorting pas.
Although an example of encoding s is shown,
Similarly for the pass, the code shown in the first embodiment may be used, or the binary symbol may be directly output as the code.

Third Embodiment In the progressive coding which is the object of the present invention, a code amount is often controlled by setting a set code amount on the coding side and stopping the coding when the set code amount is reached. The present embodiment shows an example in which the generated code amount matches the set code amount. However, since the basic configurations of the encoder and the decoder are the same as those in the first embodiment, the description thereof will be omitted.

In the encoding process, every time a codeword is determined,
It is checked whether the total number of code bits from the start of encoding has reached the set code amount, and if the determined code word is output, if it exceeds the set code amount, up to the set code amount of the code words Only output bits. For example, when the set code amount is reached in 2 bits, it is assumed that the binary symbol "0001" occurs in the context (the order is 4) in which the code word is output next. In this situation, if the fixed codeword "111" is output, the set code amount will be exceeded, so only 2 bits from the beginning of the codeword are output.

On the decoding side, it can be seen from the count value of the total number of code bits from the start of encoding that the last bit string has only 2 bits of "11", which is an incomplete code word. In this case, since it is a quaternary code, "110" or "111" can be considered as a correct code word, and "001" or "000" can be considered as a correct binary symbol candidate.
1 ". Here, since the first two bits" 00 "common to these candidates can be determined as a correct binary symbol, this is set as a reproduced binary symbol. By decoding the maximum length binary symbol from the remaining information after truncating the portion exceeding, it is possible to suppress image quality deterioration and perform accurate code amount control.

However, in some cases, even if a codeword is partially transmitted, the decoding side may not be able to decode one bit of a binary symbol. That is, the first 1 bit in the above example
If only 1 "can be sent, the correct binary symbol candidate that can be judged by the decoding side is" 1 "or"
01 ”, which means that there is no bit of a binary symbol that can be decoded.

[0063]

As described above, according to the present invention, the wavelet transforming means for wavelet transforming the image signal, and the transform coefficients from the wavelet transforming means are represented by a bit plane composed of their absolute values and positive and negative. In each of the above-mentioned bit planes, a bit plane forming means represented by a sign bit is output, and a binary symbol representing a bit in a transform coefficient equal to or less than a threshold value, which is the first pass, is output, and other transform coefficients are provided in the second pass. A binary symbol output means for outputting a binary symbol that represents a bit in, and information on which of the binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol, and the organization for the expanded information source of the binary symbol. Of the Huffman code set created dynamically, the binary syn And a binary symbol encoding means for performing binary information source encoding by selecting an optimum code for the situation of the expanded information source, a high compression rate is achieved as compared with the case of directly outputting binary symbols. The effect is that you can do it.

Further, the binary symbol encoding means converts the binary symbol output from the binary symbol output means into frequency components spatially at the same position and in the same direction in the first pass. A binary symbol sequence that represents whether or not all corresponding sets of conversion coefficients of a certain resolution or higher are below a certain threshold value, or 2 that represents whether or not a certain conversion coefficient is below a certain threshold value
A binary symbol sequence that represents the positive / negative of the conversion coefficient is classified into binary symbol sequences, and each binary sequence is independently binary symbol-encoded according to the context, and in the second pass, the binary symbol output is performed. Since the binary symbols output from the means are independently binary-source coded as one binary symbol sequence, there is an effect that the coding efficiency can be improved.

Further, the above-mentioned binary symbol coding means outputs a codeword in the procedure of generating the first coded binary symbol, and selects a certain binary symbol series from among a plurality of classified binary symbol series. Codeword is fixed, and other 2
When the codeword in the value symbol sequence is not fixed, the output order of the codeword is changed, so that only the codeword is required as the code data, and other additional information (for example, the generation procedure information of the binary symbol). ) Is not required, a higher compression rate can be achieved.

Further, the binary symbol coding means virtually inserts a dummy binary symbol even if the code word is not fixed in the last binary symbol of each path in the bit plane. Since the sign is fixed,
This has the effect of speeding up the processing.

Further, when the generated code quantity exceeds the set code quantity, the binary symbol coding means deletes all the bits exceeding the set code quantity from the generated code, so that the deterioration of the image quality is suppressed and the accuracy is improved. The code amount can be controlled.

Further, according to the present invention, based on the information indicating which of a plurality of coded binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol, it is systematic for the expanded information source of the binary symbol. And a binary symbol decoding means for performing binary information source decoding by selecting a code optimal for the situation of the extended information source of the binary symbol assumed from the estimated appearance probability of the dominant symbol from the Huffman code set created in , A first representing bits in transform coefficients below a certain threshold
Bit plane reproduction for reproducing a bit plane composed of the absolute value of the conversion coefficient and a sign bit representing positive / negative from the binary symbol of the second path and the binary symbol of the second path representing the bit in the other conversion coefficient. Means, a transform coefficient reproducing means for reconstructing a transform coefficient from a bit plane composed of the absolute value of the transform coefficient and a sign bit representing positive and negative, and a wavelet inverse transform means for performing a wavelet inverse transform on the transform coefficient. Therefore, there is an effect that a higher compression rate can be achieved as compared with the case of directly outputting a binary symbol.

In the first pass, the binary symbol decoding means determines whether all sets of transform coefficients having a certain resolution or higher corresponding to frequency components in the same spatial position and in the same direction are less than a threshold value. A binary symbol sequence indicating whether or not a certain conversion coefficient is less than a certain threshold value, a binary symbol sequence indicating whether the conversion coefficient is positive or negative, and a binary symbol sequence indicating whether the conversion coefficient is positive or negative, and each binary sequence is classified into its context. Accordingly, the binary symbols are independently decoded, and in the second pass, one binary symbol sequence is independently decoded as the binary information source, so that it is possible to improve the coding efficiency.

Further, the above-mentioned binary symbol decoding means, even if the decoded binary symbol remains without being used for reproduction of the transform coefficient at the end of each pass in the bit plane,
Since it is determined as a dummy binary symbol and the dummy binary symbol is discarded, the processing speed can be increased.

Further, the binary symbol decoding means is a code word formed by adding bits corresponding to the number of insufficient bits to the short code word when the code word located at the end of the code data is shorter than the regular code word. Since only the binary symbol part that is common to all the candidates of the decoded binary symbol sequence corresponding to is set as the decoded binary symbol, there is an effect that the deterioration of the image quality is suppressed and the accurate code amount control is possible.

Further, according to the present invention, the step of wavelet transforming the image signal, the step of expressing the transform coefficient obtained by the wavelet transform by a bit plane composed of absolute values and a sign bit representing positive or negative; In each of the bit planes, outputting a binary symbol representing a bit in a transform coefficient equal to or less than a threshold which is a first pass, and outputting a binary symbol representing a bit in another transform coefficient in a second pass. ,
From the estimated appearance probability of the dominant symbol from the Huffman code set systematically created for the extended information source of the binary symbol based on the information which one of the binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol. Since it has a step of performing the binary information source coding by selecting a code most suitable for the situation of the assumed expanded information source of the binary symbol, a higher compression rate is achieved as compared with the case of directly outputting the binary symbol. There is an effect that can be done.

Further, in the step of performing the binary information source coding, the binary symbol output in each bit plane has a resolution equal to or higher than a certain resolution corresponding to the spatially same position in the first pass. A set of conversion coefficients is classified into a binary symbol series that indicates whether or not a conversion coefficient is less than a certain threshold, a binary symbol series that indicates whether or not a certain conversion coefficient is less than a certain threshold, and a binary symbol series that indicates whether the conversion coefficient is positive or negative. , Each 2
The value sequence is independently binary-coded according to the context, and in the second pass, the binary symbols output in each bit plane are independently binary-coded as one binary symbol sequence. Therefore, there is an effect that the encoding efficiency can be improved.

Further, the step of performing the binary information source coding is such that a certain binary symbol is included in a plurality of binary symbol sequences classified so as to output a code word to the generation procedure of the head coded binary symbol. When the codeword in the sequence is fixed and the codeword in the other binary symbol sequence is not fixed, the output order of the codewords is changed, so that only the codewords are required as the code data, Since other additional information (for example, binary symbol generation procedure information) is not required,
There is an effect that a higher compression rate can be achieved.

In the step of performing the binary information source coding, a dummy binary symbol is virtually inserted even if the code word is not fixed in the last binary symbol of each path in the bit plane. As a result, the code is fixed, which has the effect of speeding up the process.

In the step of performing the binary information source coding, when the generated code amount exceeds the set code amount, all the bits exceeding the set code amount are deleted from the generated code, so that the deterioration of the image quality is suppressed. This has the effect of enabling accurate code amount control.

Further, according to the present invention, based on the information indicating which of a plurality of coded binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol, the extended information source of the binary symbol is systematically organized. From the Huffman code set created in step 1, the step of selecting a code that is optimal for the situation of the extended information source of the binary symbol assumed from the estimated appearance probability of the dominant symbol, and performing binary information source decoding, and a certain threshold value or less From the binary symbol of the first pass, which represents the bits in the transform coefficient of, and the binary symbol of the second pass, which represents the bits in the other transform coefficients, a bit plane composed of absolute values of the transform coefficients and positive / negative Reconstructing the transform coefficient from the step of reproducing the sign bit that represents and the bit plane composed of the absolute value of the transform coefficient and the sign bit that represents positive and negative And steps, since a step of inverse wavelet transform the transform coefficients, there is an effect that it is possible to achieve a high compression ratio compared to the case of outputting the binary symbol directly.

Further, the step of performing the binary information source decoding indicates whether or not all the sets of conversion coefficients of a certain resolution or higher corresponding to the same spatial position are less than a certain threshold value in the first pass. A binary symbol sequence, a binary symbol sequence that indicates whether or not a certain conversion coefficient is less than a certain threshold, and a binary symbol sequence that indicates whether the conversion coefficient is positive or negative,
Binary information source decoding is independently performed on the value sequence according to the context, and in the second pass, one binary symbol sequence is independently binary information source decoding, thus improving the coding efficiency. The effect is that you can.

In the step of performing the binary information source decoding, even if the decoded binary symbol remains without being used for the reproduction of the transform coefficient at the end of each pass in the bit plane, the dummy binary symbol is used. 2 of the dummy
Since the value symbol is discarded, there is an effect that the processing speed can be increased.

In the step of performing the binary information source decoding, when the code word located at the end of the code data is shorter than the regular code word, the short code word is added with bits corresponding to the number of insufficient bits. Since only the binary symbol portion common to all the candidates of the decoded binary symbol sequence corresponding to the possible codeword is used as the decoded binary symbol, there is an effect that the deterioration of the image quality is suppressed and the accurate code amount control is possible. .

[Brief description of drawings]

FIG. 1 is a coding block diagram showing a first embodiment of the present invention.

FIG. 2 is a decoding block diagram showing the first embodiment of the present invention.

FIG. 3 is a diagram for explaining entropy coding according to the first embodiment of the present invention.

FIG. 4 is a diagram showing a state transition table in the first embodiment of the present invention.

FIG. 5 is a diagram showing an example of significant / insignificant coefficients in a bit plane according to the first embodiment of the present invention.

FIG. 6 is a diagram showing an example of binary symbol output according to the second embodiment of the present invention.

FIG. 7 is a diagram showing an example of binary symbol output according to the second embodiment of the present invention.

FIG. 8 is a diagram showing an example of specific binary symbol output according to the second embodiment of the present invention.

FIG. 9 is a diagram showing conversion coefficients represented by a general tree structure.

FIG. 10 is a flowchart showing a conventional encoding process.

FIG. 11 is a flowchart showing a conventional encoding process.

FIG. 12 is a flowchart showing a conventional encoding process.

[Explanation of symbols]

701: Wavelet transform unit, 702 bit plane conversion unit, 703 binary symbol output unit, 704 binary symbol encoding unit, 801 binary symbol decoding unit, 802
Bit plane reproduction unit, 803 Conversion coefficient reproduction unit, 8
04 Wavelet inverse transform unit.

   ─────────────────────────────────────────────────── ─── Continued front page    F-term (reference) 5C053 FA20 GB22 GB26 LA14                 5C059 KK01 KK11 KK22 MA24 MA35                       MC31 ME02 TA49 TB14 TD10                       UA02 UA05                 5C078 BA53 CA01 DA01 DA02 DA21                 5J064 AA01 BA16 BB13 BC23 BD03

Claims (18)

[Claims] 1. A wavelet transforming means for wavelet transforming an image signal, a bit plane converting means for expressing transform coefficients from the wavelet transforming means with a bit plane composed of absolute values thereof and with a sign bit representing positive / negative, A binary value that outputs a binary symbol that represents a bit in a transform coefficient equal to or less than a threshold value that is a first pass in each bit plane, and outputs a binary symbol that represents a bit in another transform coefficient in a second pass From the Huffman code set systematically created for the extended information source of the binary symbol, based on the symbol output means, information on which of the binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol, the dominant symbol Select the most suitable code for the situation of the expanded information source of the binary symbol assumed from the estimated appearance probability of the symbol. An image coding apparatus comprising: a binary symbol coding means for selectively coding a binary information source. 2. The binary symbol encoding means comprises the
In the first pass, the binary symbols output from the value symbol output means are all set of transform coefficients having a resolution equal to or higher than a certain resolution and corresponding to frequency components in the same spatial position and in the same direction. A binary symbol sequence indicating whether or not a certain conversion coefficient is less than a certain threshold value, a binary symbol sequence indicating whether the conversion coefficient is positive or negative, and a binary symbol sequence indicating whether the conversion coefficient is positive or negative, and each binary sequence is classified into its context. According to the second pass, the binary symbols are independently encoded, and in the second pass, the binary symbols output from the binary symbol output means are independently binary source coded as one binary symbol sequence. The image coding apparatus according to claim 1. 3. The binary symbol encoding means determines a codeword in a certain binary symbol sequence among a plurality of classified binary symbol sequences, and codewords in other binary symbol sequences. 3. The image coding apparatus according to claim 1, wherein the output order of the code words is changed when is not determined. 4. The binary symbol encoding means virtually inserts a dummy binary symbol even if the code word is not fixed in the last binary symbol of each path in the bit plane. The image coding apparatus according to claim 1, wherein the code is fixed. 5. The binary symbol encoding means, when the generated code amount exceeds a set code amount, deletes all the bits exceeding the set code amount from the generated code. 4. The image encoding device according to any one of 4 above. 6. A Huffman code set systematically created for the extended information source of the binary symbol based on information indicating which of a plurality of encoded binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol. , A binary symbol decoding means for performing binary information source decoding by selecting an optimum code for the situation of the extended information source of the binary symbol assumed from the estimated appearance probability of the dominant symbol, and conversion below a certain threshold value. From the binary symbol of the first pass, which represents the bit in the coefficient, and the binary symbol of the second path, which represents the bit in the other transform coefficient, a bit plane composed of the absolute value of the transform coefficient and a sine sign representing positive and negative Reconstructing the transform coefficient from the bit plane playback means that plays back the bit, the bit plane that is composed of the absolute value of the above transform coefficient, and the sign bit that represents the sign A conversion coefficient reproduction unit that, the image decoding apparatus being characterized in that a wavelet inverse transform means for inverse wavelet transform the transform coefficients. 7. The binary symbol decoding means comprises the first
In the path of, the binary symbol sequence indicating whether or not all the sets of conversion coefficients of a certain resolution or more corresponding to frequency components in the same spatial position and in the same direction are less than a certain threshold value, a certain conversion coefficient is less than a certain threshold value In this second pass, the binary symbol sequence representing whether or not the binary coefficient symbol representing the positive or negative of the transform coefficient is classified, and the binary symbol sequence is independently decoded according to the context. Is
7. The image decoding apparatus according to claim 6, wherein one binary symbol sequence is independently decoded by a binary information source. 8. The binary symbol decoding means determines as a dummy binary symbol even if the decoded binary symbol remains at the end of each pass in the bit plane without being used for reproduction of transform coefficients. The image decoding device according to claim 6 or 7, wherein the dummy binary symbols are discarded. 9. The code word formed by the binary symbol decoding means, when the code word located at the end of the code data is shorter than the normal code word, by adding bits for the number of insufficient bits to the short code word. 9. The image decoding apparatus according to claim 6, wherein only a binary symbol portion common to all of the decoded binary symbol sequence candidates corresponding to is set as a decoded binary symbol. 10. A step of wavelet transforming an image signal, a step of expressing a transform coefficient obtained by the wavelet transform by a bit plane composed of absolute values and a sign bit representing positive / negative, and in each of the bit planes, Outputting a binary symbol representing a bit in a conversion coefficient equal to or less than a threshold which is a first pass, and outputting a binary symbol representing a bit in another conversion coefficient in a second pass; Based on the information indicating which is the dominant symbol and the estimated appearance probability of the dominant symbol, a binary value assumed from the estimated occurrence probability of the dominant symbol from the Huffman code set systematically created for the extended information source of the binary symbol. Step of performing binary information source coding by selecting the most suitable code for the situation of expanded information source of symbol An image encoding method comprising: 11. The step of performing the binary information source coding, the binary symbol output in each bit plane is equal to or higher than a certain resolution corresponding to the same spatial position in the first pass. A set of conversion coefficients is classified into a binary symbol series that indicates whether or not a conversion coefficient is less than a certain threshold, a binary symbol series that indicates whether or not a certain conversion coefficient is less than a certain threshold, and a binary symbol series that indicates whether the conversion coefficient is positive or negative. , Each 2
The value sequence is independently binary-coded according to the context, and in the second pass, the binary symbols output in each bit plane are independently binary-coded as one binary symbol sequence. The image coding method according to claim 10, wherein: 12. The step of performing binary information source coding, wherein a codeword in a certain binary symbol sequence is determined in a plurality of classified binary symbol sequences, and another binary symbol sequence is used. 12. The image coding method according to claim 10, wherein the output order of the codewords is changed when the codewords of are not determined. 13. The step of performing the binary information source coding virtually inserts a dummy binary symbol even if the code word is not fixed in the last binary symbol of each path in the bit plane. The image coding method according to any one of claims 10 to 12, wherein the code is determined by the above. 14. The binary information source encoding step is characterized in that, when the generated code amount exceeds a set code amount, all bits exceeding the set code amount are deleted from the generated code. The image coding method according to any one of Items 10 to 13. 15. A Huffman code set systematically created for the extended information source of the binary symbol based on information indicating which of a plurality of encoded binary symbols is the dominant symbol and the estimated appearance probability of the dominant symbol. From among the above, the step of selecting the optimum code for the situation of the extended information source of the binary symbol assumed from the estimated appearance probability of the dominant symbol and performing the binary information source decoding, and the bit in the transform coefficient below a certain threshold From the binary symbol of the first pass that represents and the binary symbol of the second pass that represents the bits in the other transform coefficients, a bit plane composed of the absolute value of the transform coefficient and a sign bit that represents positive or negative are reproduced. And a step of reconstructing a transform coefficient from a bit plane composed of absolute values of the transform coefficient and a sign bit representing positive and negative, And a step of performing inverse wavelet transform on the coefficient. 16. The step of performing the binary information source decoding indicates, in the first pass, whether or not all sets of transform coefficients corresponding to a spatially same position and having a certain resolution or more are less than a certain threshold value. A binary symbol sequence, a binary symbol sequence that indicates whether or not a certain conversion coefficient is less than a certain threshold, and a binary symbol sequence that indicates whether the conversion coefficient is positive or negative,
16. The binary information source decoding is independently performed according to the context of the value sequence, and the binary information source decoding is independently performed for one binary symbol sequence in the second pass. Image coding method. 17. The step of performing the binary information source decoding is a dummy binary symbol even if the decoded binary symbol remains at the end of each pass in the bit plane without being used for reproduction of the transform coefficient. 2 of the dummy
The image decoding method according to claim 15, wherein the value symbol is discarded. 18. The step of performing the binary information source decoding, when the code word located at the end of the code data is shorter than the regular code word, adds bits corresponding to the number of insufficient bits to the short code word. 2. Only the binary symbol part common to all the candidates of the decoded binary symbol sequence corresponding to the possible codeword is set as the decoded binary symbol.
The image decoding method according to any one of 5 to 17.