JPH10229340A - Coding and decoding system for sub-band signal and wavelet transformation coefficient - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve coding efficiency by selecting adaptively a scanning pattern, so as to maximize a consecutive skip number of a specific coefficient in the case of operating a set of grouped coefficients to be a linear coefficient string. SOLUTION: This system is provided with a scanning pattern estimate circuit 604, a scanning converter 503, a specific pattern replacement circuit 606, and a coder 605. The grouped quantization transformation coefficient is given to the scanning pattern estimate circuit 604, in which a proper scanning pattern is estimated. A scanning converter 503 scans the grouped transformation coefficient, according to the scanning pattern and generates a linear transformation coefficient series. The consecutive specific coefficient series at the end in the generated transformation coefficient is replaced with a transformation coefficient, including a scanning stop code by the specific pattern replacement circuit 606, and a coder 605 converts the replaced string into a code string for transmission or storage. Estimation of the scanning pattern is conducted for each frequency band, from a low-frequency pattern to a high-frequency pattern.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to high efficiency coding of digital signals.

[0002]

2. Description of the Related Art As a high-efficiency encoding technique for digital signals such as images, there are a sub-band encoding scheme for dividing a signal into bands and encoding the signals, and a wavelet transform encoding scheme. In these systems, a signal obtained as a result of being decomposed by a subband or a wavelet filter (hereinafter referred to as a transform coefficient) is quantized and then replaced with a one-dimensional code sequence to thereby obtain a highly compressed code. Perform the conversion. As a method of obtaining a code sequence from the quantized transform coefficients, in addition to a method of converting a transform coefficient into a code sequence for each decomposed band, a specific portion of a transform coefficient is cut out to form a group,
There is a method in which each group is independently converted into a code string. As an example of the latter, an example in which an image signal is encoded is described in "Hybrid Picture Coding by Ota et al.
withWavelet Transform an
d Overlapped Motion-Compe
nsated Interframe Predict
Ion Coding ", December 1993, IEE Transactions on Signal Processing, Vol. 41, No. 12, No. 3416-34.
Page 24 (IEEE Transactions on
Signal Processing, vol. 41,
no. 12, December, 1993) and Sha
“Embedded Image Co
Ding Use Zerotrees of W
avelet Coefficients ", 1993
December, IEE Transactions on Signal Processing, Vol. 41, No. 1
2, 3445-3462 (IEEE Transac)
Tions on Signal Processin
g, vol. 41, no. 12, December, 1
993), and “Description of c.
ore experiments on coding
efficiency in MPEG-4 Vid
eo ", September 1996, I S / O / I / C / J / C / 1 / S / C / 29 /
WJ11, N1385 (ISO / IEC
JTC1 / SC29 / WG11 N1385, Sept
Ember, 1996) "Wavelet Codi.
ng of I and P Pictures, Ap
pendix C.I. Description of T
ree StructureRunlength Co
ding ".

In these examples, a transform coefficient is divided into square sections for each band, and sections of each band corresponding to the same position in space are collected and processed as a transform coefficient group for each group. Each method focuses on a large number of 0 coefficients among the transform coefficients in the group, and by adding information indicating the position, skips these 0 coefficients without encoding, and as a result, This achieves a substantial reduction in the amount of code.

FIG. 12 shows a method of generating transform coefficients used in these methods. The original image is first divided into four by a filter bank composed of four filters, filters 1 to 4, and then by a sub-sampling circuit 17 to 20 having the same configuration, both horizontally and vertically at a ratio of 2: 1. Interval thinning is performed. Here, the filters 1 to 4 are filters that divide the input signal into four according to the spatial frequency, and the ranges corresponding to # 1 to # 4 on the spatial frequency division diagram shown in FIG. Of the signals obtained by the thinning, only the output of the sampling circuit 20, which is a signal corresponding to the band of # 1 in FIG. 13, is similarly divided into four using the filters 5 to 8 and the sub-sampling circuits 21 to 24. I do. Further, the output of the sub-sampling circuit 24 is
The signal is again divided into four by the 12 and sub-sampling circuits 25 to 28. Do the same thing again with filters 13-1.
6 and the sub-sampling circuits 29 to 32, and the band division is completed. Here, the filters 5 to 8 are the same as the filters 1 to 4, respectively. Filters 9 to 12 and 13 to 16 are similarly set to filter 1
To 4, respectively. The sub-sampling circuits 17 to 22 are all identical circuits for thinning out digital signals at equal intervals both horizontally and vertically at a ratio of 2: 1. Hereinafter, the signals of a total of 13 channels obtained in FIG. 12 are distinguished by the subgroup number and the layer number. The subgroup number refers to the number of the pass band of the filter that has passed last, and the layer number refers to the total number of filters that have passed.

The signal of each channel has a different magnitude depending on the layer for subsampling.
Assuming that the size of the original image is W × H, the size W × H of the n-th layer is

[0006]

(Equation 1)

[0007] In order to extract the (i, j) -th (i, j = 1, 2,..., W / B) square section of size B × B from each channel, assuming that the layer number is n,

[0008]

[Outside 1]

[0009] This can be realized by taking out the area. In this method, a group of change coefficients belonging to a certain square block obtained by the above procedure is further classified by subgroup number, and a final grouped transform coefficient is obtained.

FIG. 14 is a block diagram of the scanning and encoding unit of the encoder of the system described in the third document mentioned above. The scan converter 503 scans the transform coefficients grouped by the above-described method using a predetermined scan pattern to generate a one-dimensional transform coefficient sequence. The scanning pattern is different for each subgroup number,
In the case of this system adopting B = 16, four types shown in FIG. 15 are assigned in correspondence with the subgroup numbers. The specific pattern replacement circuit 606 deletes the continuous 0 coefficient at the end of the conversion coefficient sequence by replacing it with the scan abort code. The transform coefficient sequence including the obtained scan truncation code is converted into a code sequence in the encoder 605. At this time, the transform obtained by using the zero run length indicating the length of the continuous 0 coefficient as additional information is performed. By skipping all 0 coefficients in the coefficient sequence, the final code amount is suppressed. FIG. 16 is a block diagram of the reverse scanning and decoding unit of the decoder of this system. Here, the inverse operation of the encoder 605 is performed, and the decoder 804 converts the code sequence output from the encoder to a transform coefficient sequence including a scan truncation code. The scan abort code replacement circuit 805 replaces the scan abort code of the transform coefficient sequence including the input scan abort code with 0 coefficients that continue to the end to generate a transform coefficient sequence. Finally, the inverse scan converter 703 performs inverse processing of the scan converter 503 in the encoder, and outputs grouped transform coefficients.

In the decoder of this system, the transformed coefficients obtained by the above processing are combined for all the groups, and then the inverse wavelet transform is performed to obtain a final decoded image.

[0012]

In the case of introducing interlaced scanning of a specific coefficient such as the 0 coefficient, it is desirable that the specific coefficients to be skipped are as continuous as possible, because the code reduction effect is large. However, since the processing (scanning) order of the transform coefficients is fixed in any of the examples given as the prior art, the continuity of the specific coefficients is not pursued to the utmost, and as a result, there is room for improving the coding efficiency. There is.

[0013]

According to the present invention, when a set of grouped coefficients is scanned into a one-dimensional coefficient sequence, the number of continuous readings of a specific coefficient is skipped by adaptively switching a scanning pattern. By doing so, the above-mentioned problem is solved.

In the first aspect of the present invention, when sequentially scanning the transform coefficients in the group for each layer, a plurality of scan patterns are prepared for each layer, and a continuous specific coefficient to be skipped from the plurality of scan patterns is prepared. It is characterized by comprising means for selecting a scan pattern as many as possible, means for executing a scan in accordance with the pattern to generate a transform coefficient sequence, and means for encoding the transform coefficient sequence and the scan pattern.

According to a second aspect of the present invention, in the first aspect,
In the invention of (1), the continuous specific coefficients existing at the end of the transform coefficient sequence are replaced with one scanning cutoff code. By selecting the scanning pattern so that the total number of coefficients deleted by this replacement becomes the maximum, the final reduction of the code amount can be realized without estimating the actual code amount when selecting the scan pattern. For this purpose, in addition to the first aspect of the present invention, the apparatus further comprises means for replacing a specific coefficient continuous at the end of the transform coefficient sequence with one scanning abort code.

According to a third aspect of the present invention, there is provided the first aspect.
Means for decoding the code data and separating and extracting the conversion coefficient sequence and the scanning pattern, and converting the conversion coefficient sequence according to the extracted scanning pattern of each layer in order to decode the code data output from the invention of It is characterized by comprising means for converting into coefficients.

According to a fourth aspect of the present invention, in the second aspect,
In order to decode the code data output from the invention of the third aspect, in addition to the third decoding method of the invention, a means for replacing the scan abort code with a continuous number of specific coefficient sequences is provided.

According to a fifth aspect of the present invention, there is provided the first aspect.
In order to improve the coding efficiency by reducing the scanning pattern of each layer, which is additional information required in the invention of the present invention, the lower layer corresponding to the high frequency component is effectively converted from the coefficient of the upper layer corresponding to the low frequency component. Estimate the scanning pattern to be scanned. For this purpose, a means for estimating an effective lower layer scanning pattern from an upper layer, a means for sequentially scanning transform coefficients in a group from an upper layer according to the estimated scanning pattern to generate a transform coefficient sequence, It is characterized by comprising means for coding only the coefficient sequence.

According to a sixth aspect of the present invention, in the second aspect,
In order to reduce the scanning pattern, which is additional information necessary in the invention of the present invention, and to improve the coding efficiency, the scanning that effectively scans the lower layer corresponding to the high frequency component from the coefficient of the upper layer corresponding to the low frequency component Estimate the pattern. For this purpose, in addition to the second aspect of the present invention, there is provided means for estimating an effective lower layer scanning pattern from an upper layer, instead of a circuit which uses the transformed coefficient sequence and the scanning pattern as codes after replacement. And means for coding only the transform coefficient sequence replaced with.

According to a seventh aspect of the present invention, in the fifth aspect,
Means for decoding the code data to obtain a transform coefficient sequence in order to decode the code data output from the invention according to claim 5, and from the transform coefficients of the upper layer decoded using the same algorithm as in the fifth invention. It is characterized by comprising means for determining a scanning pattern of a lower layer, and means for converting a conversion coefficient sequence into conversion coefficients according to the determined scanning pattern of each layer.

According to an eighth aspect of the present invention, in the sixth aspect,
In order to decode the code data outputted from the invention of the seventh aspect, in addition to the decoding method of the seventh aspect of the present invention, the apparatus further comprises means for replacing the scan abort code with a plurality of continuous specific coefficient sequences.

[0022]

FIG. 1 is a block diagram showing an embodiment of the first invention. This is a circuit for scanning and encoding the grouped transform coefficients for each frequency band. The grouped conversion coefficients are input to the scan pattern determination circuit 104 and the scan converter 103. The scan pattern output from the scan pattern determination circuit 104 is input to the scan converter 103. The scan converter 103 outputs a conversion coefficient sequence. The output transform coefficients are input to the encoder 105. A code string is output from the encoder 105.

Next, the operation of each component will be described. The scan pattern determination circuit 104 determines and outputs a scan pattern of a transform coefficient for reducing the code amount based on the grouped transform coefficients. The details will be described later.
The scan converter 103 scans the grouped conversion coefficients based on the input scanning pattern, and generates a one-dimensional conversion coefficient sequence. The encoder 105 encodes the input transform coefficient sequence and the scanning pattern, and outputs a code sequence.

Next, the scan converter 103 will be described in detail. Here, a scanning pattern is determined from the grouped transform coefficients. Here, the scanning pattern is data that defines the order of scanning the grouped transform coefficients. First, the grouped transform coefficients are separated for each layer, and then the scanning pattern is determined independently for each layer. The scanning pattern determination algorithm for each layer will be described with reference to the flowchart in FIG.

First, the counter n indicating the number of the scanning pattern is set to 0, the minimum value e _min of the estimated code amount e is set to a sufficiently large value, and the value n of the counter n when the estimated code amount becomes the minimum is set. Set _min to 0 (step 9
00). Next, the conversion coefficients grouped by the predetermined n-th scan pattern are scan-converted (step 901). Then, the estimated code amount e at this time is calculated (step 902). The estimated code amount will be described later. Here, the obtained estimated code amount e is compared with the minimum value e _min of the estimated code amount already obtained (step 903). At this time, the estimated code amount e becomes e
If it is smaller than _min, the value of the estimated code amount e is set to e
substituted in _min, and the value of the counter n at this time is substituted into n _min (step 904). On the other hand, if the estimated code amount e is larger than e _min , step 904 is skipped. Next, the value of the counter n is compared with the total number N of scanning patterns (step 905). If n is smaller than N, the value of the counter n is increased by 1 (step 907), and the process returns to step 901. On the other hand, if n is not smaller than N, the n _min- th scanning pattern is output (step 906), and the series of procedures ends.

Here, the method of calculating the estimated code amount e includes:
There are several possibilities. In the first method, the occurrence probability of a symbol to be encoded such as a transform coefficient is determined in advance, and the estimated code amount is calculated using the occurrence probability.
The second method is a first method in which the symbol occurrence probability is gradually updated according to the frequency of occurrence of a symbol to be encoded. This method has an advantage that the code amount can be more accurately estimated because the occurrence probability is adaptively changed according to the appearance frequency. The third method is to feed back the actual code amount from the encoder 105 in FIG. 1 and use it instead of the estimated code amount. This has an advantage that a scanning pattern for suppressing the code amount can be selected more accurately than the first and second methods.

The transform coefficient sequence output from the scan converter 103 and the scan pattern output from the scan pattern determination circuit 104 are encoded by an encoder 105. Here, as for the scanning pattern, the scanning pattern itself may be encoded, or the same scanning pattern as the encoding side is prepared on the decoding side, and only the index for specifying the scanning pattern is encoded. You may make it. Further, a system combining these two is also possible. That is, the scanning pattern that is frequently selected is also prepared on the decoding side, and only the index is encoded. For the other scanning patterns, the scanning pattern itself is encoded. Further, a method is also possible in which the scanning patterns used for decoding are sequentially registered on the decoding side, and when the next scanning pattern occurs, only the index for specifying the scanning pattern is encoded.

As described above, by changing the scanning pattern adaptively for each layer in accordance with the contents of the blocked conversion coefficients, the coding efficiency can be improved as compared with the conventional method using a fixed scanning pattern. .

FIG. 2 is a block diagram showing an embodiment of the second invention. This is a circuit for scanning and encoding the grouped transform coefficients for each frequency band. The grouped quantized transform coefficients are input to the scan pattern determination circuit 204 and the scan converter 103. The scan pattern output from the scan pattern determination circuit 204 is input to the scan converter 103. From the scan converter 103,
A conversion coefficient sequence is output. The output conversion coefficient sequence is input to the specific pattern replacement circuit 206. A conversion coefficient sequence including the scan abort code is output from the specific pattern replacement circuit 206 and input to the encoder 205. A code sequence is output from the encoder 205.

Next, the operation of each component will be described. The scan pattern estimating circuit 204 determines and outputs a scan pattern of the transform coefficient for reducing the code amount based on the grouped transform coefficients. The scan converter 103 is
Based on the input scanning pattern, the grouped conversion coefficients are scanned to generate a one-dimensional conversion coefficient sequence. The specific pattern replacement circuit 206 replaces the continuous coefficient with a scan abort code when a specific value continues to the end of the grouped range in the conversion coefficient sequence output from the scan converter 103. , And outputs a transform coefficient sequence including the scan truncation code. The encoder 205 encodes the transform coefficient sequence including the scan abort code and the scan pattern, and outputs a code sequence.

Next, the operation of the scanning pattern determination circuit 204 will be described in detail. The basic operation of the scanning pattern determining circuit 204 is the same as that of the scanning pattern determining circuit 104 in FIG. Therefore, the same circuit as the scanning pattern determination circuit 104 in FIG. 1 can be used. In addition to this, instead of estimating the code amount itself, this circuit uses the number of transform coefficients replaced by the scan truncation code,
A circuit for determining a scanning pattern can also be used. In this case, the scanning pattern is determined so as to increase the number as much as possible. This is because the code amount can be reduced by increasing the number of transform coefficients to be replaced. For example,
When the encoder 105 encodes the length of the zero run, the scanning pattern may be determined so that the 0 coefficient is continued at the end of the group as much as possible. Thereby, the scanning pattern determination circuit can be simplified.

FIG. 3 is a block diagram showing an embodiment of the third invention. This is a circuit that decodes the code string created in the first invention to obtain a group of transform coefficients that can be subband combined or inverse wavelet transformed. First,
The code sequence is input to the decoder 304. The decoder 304 outputs a conversion coefficient sequence and a scanning pattern, and both are input to the inverse scanning converter 303. The inverse scanning converter 303 outputs the grouped conversion coefficients.

Next, the operation of each component will be described. The decoder 304 decodes the input code sequence, separates the code sequence into a transform coefficient sequence and a scanning pattern, and outputs each of them. If an index of a predetermined scanning pattern is encoded instead of the scanning pattern, a scanning pattern corresponding to the index is output. The inverse scan converter 303 restores the one-dimensional conversion coefficient sequence to a state before scanning according to the input scanning pattern, and outputs it as grouped conversion coefficients.

According to the third aspect, the group of transform coefficients encoded in the first aspect can be completely restored.

FIG. 4 is a block diagram showing an embodiment of the fourth invention. This is a circuit that decodes the code string created in the second invention to obtain a group of transform coefficients that can be sub-band synthesized or inverse wavelet transformed. First, the code string is input to the decoder 404. Decoder 404
Outputs a conversion coefficient sequence including a scan abort code and a scan pattern. The conversion coefficient sequence including the scan abort code is input to the scan abort code replacement circuit 405, and the scan pattern is input to the inverse scan converter 303. A conversion coefficient sequence is output from the scanning abort code replacement circuit 405,
It is input to the inverse scanning converter 303. Inverse scanning converter 303
Output the grouped transform coefficients.

Next, the operation of each component will be described. The decoder 404 decodes the input code sequence, separates the code sequence into a transform coefficient sequence including a scan abort code, and a scan pattern, and outputs each of them. If a predetermined scan pattern index is encoded instead of the scan pattern, a scan pattern corresponding to this index is output. Scan truncation code replacement circuit 405
Inserts a specific value to the end of the group in the scan abort code and outputs it as a transform coefficient sequence. The inverse scan converter 303 restores the one-dimensional conversion coefficient sequence to a state before scanning according to the input scanning pattern, and outputs the converted conversion coefficient sequence as grouped conversion coefficients.

According to the fourth aspect, the group of transform coefficients encoded by the second aspect can be completely restored.

FIG. 5 is a block diagram showing an embodiment of the fifth invention. This is a circuit for sequentially scanning and encoding the grouped quantized transform coefficients from a low frequency band to a high frequency band. The grouped quantized transform coefficients are input to the scan pattern estimation circuit 504 and the scan converter 503. The scan pattern output from the scan pattern estimation circuit 504 is input to the scan converter 503. The scan converter 503 outputs a conversion coefficient sequence. The output transform coefficient sequence is input to the encoder 505. The encoder 505 outputs a code string.

Next, the operation of each component will be described. The scanning pattern estimating circuit 504 estimates and outputs the scanning pattern of each layer from the grouped quantized transform coefficients so as to reduce the code amount. The details will be described later. The scan converter 503 sequentially scans the grouped quantized conversion coefficients from a low-frequency band to a high-frequency band based on the input scanning pattern, and generates a one-dimensional conversion coefficient sequence. The encoder 505 encodes the input transform coefficient sequence 505 and outputs a code sequence.

Next, details of the scanning pattern estimation circuit 504 will be described with reference to FIG. First, the input grouped quantized transform coefficients are input to a layer separation circuit 2000.
Is separated into L layers. The conversion coefficient of each separated layer is input to a scanning pattern determination circuit for each layer. The m-th (m = 1,..., L−1) scan pattern determination circuit determines the scan pattern of the layer m from the quantized transform coefficients of the layers m + 1,. The scanning patterns of the respective layers obtained by the first to L-1th scanning pattern estimating circuits are all
03 is input.

The m-th scanning pattern determination circuit determines a scanning pattern in which the final code amount is small by using the conversion coefficients of the layers m + 1 to L. For example, when the length of the zero run is encoded by the encoder 505 in FIG. 5, the distribution of 0 coefficients in the layer m is calculated from the transform coefficients of the layers m + 1 to L using the correlation of the transform coefficients between the layers. Predict. Then, the scan pattern of the conversion coefficient is determined so that the zero run continues for a long time.

The following method can be considered as an embodiment for realizing this by making it easy to continue the zero run at the end of each layer. First, several scanning patterns are prepared. For example, the sub-blocks # 2, # 3,
For # 4, (a), (b),
A scanning pattern as shown in (c) is prepared. (A),
There are four scanning patterns in each of (b) and (c), but these four have different scanning start positions. The scanning direction is along the direction in which the values of the transform coefficients are correlated in each of the sub-blocks # 2, # 3, and # 4. Next, a power distribution of the conversion coefficient in the upper layer is obtained, and a desired scanning pattern is selected from among the four patterns by using the correlation of the conversion coefficient between the layers. For example, for the four corners of the transform coefficient block in the upper layer, the power around the four corners is calculated, and a scanning pattern starting from the corner where the power is maximum is selected.

In addition to this, there are a plurality of scan pattern determination methods. For example, there are a method of outputting a scan pattern that scans an area having the lowest power last, and a method of outputting a scan pattern that scans from a corner having the highest power to a corner having the next highest power. Instead of using power, it is also possible to use statistics such as the sum of absolute values of transform coefficients and the appearance frequency of non-zero coefficients. Various scanning patterns other than those shown in FIG. 11 are conceivable.

According to the scanning pattern thus obtained, the conversion coefficient is scanned from the layer L to the layer 1 and encoded. On the decoding side, the code string obtained in this way can be completely restored by restoring the transform coefficients from layer L to layer 1.

The fifth invention has the advantage that the amount of code can be reduced as compared with the conventional system using a fixed scan pattern without coding the information of the scan pattern.

FIG. 6 is a block diagram showing an embodiment of the sixth invention. This is a circuit for sequentially scanning and encoding the grouped quantized transform coefficients from a low frequency band to a high frequency band. The grouped quantized transform coefficients are input to the scan pattern estimation circuit 604 and the scan converter 503. The scan pattern output from the scan pattern estimation circuit 604 is input to the scan converter 503. From the scan converter 503, a conversion coefficient sequence is output and input to the specific pattern replacement circuit 606. The conversion coefficient sequence including the scan abort code is output from the specific pattern replacement circuit 606 and input to the encoder 605. Encoder 60
5 outputs a code string.

Next, the operation of each component will be described. The scanning pattern estimating circuit 604 estimates and outputs a scanning pattern of each layer to reduce the code amount from the grouped quantized transform coefficients. The scan converter 503 sequentially scans the grouped quantized transform coefficients from a low-frequency band to a high-frequency band based on the input scan pattern, and generates a one-dimensional transform coefficient sequence. The specific pattern replacement circuit 606 replaces the continuous coefficient with a scan abort code when a specific value continues to the end of the grouped range in the conversion coefficient sequence output from the scan converter 503, A transform coefficient sequence including a scan abort code is output. The encoder 605 encodes the transform coefficient sequence including the scan truncation code, and outputs a code sequence.

Next, the operation of the scanning pattern estimation circuit 604 will be described in detail. The basic operation of the scan pattern estimating circuit 604 is the same as that of the scan pattern estimating circuit 504 of FIG. 5, and is shown in FIG. The operation of the m-th (m = 1,..., L−1) scanning pattern determination circuit may be the same as that of the scanning pattern estimation circuit 504 in FIG. However, instead of estimating a scan pattern that reduces the code amount itself, a circuit that estimates a scan pattern that increases the number of transform coefficients replaced by the scan truncation code may be used. This is done by increasing the number of transform coefficients
This is because the code amount can be reduced. For example, the encoder 60
In step 5, when encoding the length of the zero run, it is sufficient to estimate a scanning pattern in which a zero coefficient is continued as much as possible at the end of the group.

The sixth aspect of the invention has the advantage that the amount of code can be reduced as compared with the conventional method using a fixed scan pattern without coding the information of the scan pattern. FIG. 7 is a block diagram showing an embodiment of the seventh invention. This is a circuit that decodes the code sequence created in the fifth invention to obtain a group of transform coefficients that can be subband combined or inverse wavelet transformed. First, the code string is decoded by the decoder 704.
Is input to A transform coefficient sequence is output from the decoder 704 and input to the inverse scan converter 703. The inverse scan converter 703 outputs grouped conversion coefficients. The grouped transform coefficients are input to the scanning pattern estimation circuit 705. Scan pattern estimation circuit 70
5, a scanning pattern is output, and the inverse scanning converter 7
03 is output.

Next, the operation of each component will be described. Decoder 704 decodes the input code sequence and outputs a transform coefficient sequence. The inverse scan converter 703 sequentially restores the low-frequency band to the high-frequency band according to the input scan pattern, and outputs a group of coefficients that can be subband synthesized or inverse wavelet transformed. The scanning pattern estimating circuit 705 is the same circuit as the scanning pattern estimating circuit 504 used in the encoder of FIG. 5, and estimates and outputs a scanning pattern from the grouped transform coefficients after restoration.
This configuration is as shown in FIG. However, FIG.
The layer separation circuit 2000 has a storage device, and the restored conversion coefficients are held therein until all the conversion coefficients in the group are restored.

According to the seventh aspect, the group of the transform coefficients encoded in the fifth aspect can be completely restored.

FIG. 8 is a block diagram showing an embodiment of the eighth invention. This is a circuit that decodes the code string created in the sixth invention to obtain grouped transform coefficients. First, the code string is input to the decoder 804. From the decoder 804, a transform coefficient sequence including the scan abort code is output and input to the scan abort code replacement circuit 805. A conversion coefficient sequence is output from the scan abort code replacement circuit 805 and input to the inverse scan converter 703. The inverse scan converter 703 outputs grouped conversion coefficients. The grouped transform coefficients are input to the scan pattern estimation circuit 806. The scanning pattern is output from the scanning pattern estimation circuit 806 to the inverse scanning converter 703.

Next, the operation of each component will be described. The decoder 804 decodes the input code sequence and outputs a transform coefficient sequence including a scan truncation code. The scan abort code replacement circuit 805 adds
Insert a specific value to the end of the group and output it as a transform coefficient sequence. The inverse scan converter 703 sequentially restores the low-frequency band to the high-frequency band according to the input scan pattern, and outputs a group of coefficients that can be subband synthesized or inverse wavelet transformed. Scan pattern estimation circuit 80
6 is the same as the scan pattern estimation circuit 604 used in the encoder of FIG. 6, and estimates and outputs a scan pattern from the grouped transform coefficients after restoration. However, FIG.
0 has a storage device in the layer separation circuit 2000,
Until all transform coefficients in the group are restored,
It is assumed that the restored conversion coefficient is held.

According to the eighth aspect, the group of the transform coefficients encoded in the sixth aspect can be completely restored.

[0055]

By combining the first encoder of the present invention with the third decoder of the present invention, or combining the second encoder of the present invention and the fourth decoder of the present invention, it is possible to read at a time. The length of continuous specific coefficients that can be skipped can be lengthened, and coding efficiency can be improved. As an example, Table 1 shows how the code amount can be reduced as compared with the conventional method when the second encoder and the fourth decoder of the present invention are used and 1-bit additional information is given to each group. .

[0056]

[Table 1]

By combining the fifth encoder of the present invention and the seventh decoder, or combining the sixth encoder of the present invention and the eighth decoder of the present invention, no additional information is added. The length of continuous specific coefficients that can be skipped at a time can be lengthened, and coding efficiency can be improved. As an example, the sixth encoder of the present invention and the eighth encoder of the present invention
Layer above (high-frequency band)
Table 2 shows how the amount of code can be reduced in comparison with the conventional method when a method of starting scanning in a predetermined direction from a corner having the maximum local power is used.

[0058]

[Table 2]

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a first encoder according to the present invention.

FIG. 2 is a configuration diagram of a second encoder according to the present invention.

FIG. 3 is a configuration diagram of a third decoder of the present invention.

FIG. 4 is a configuration diagram of a fourth decoder according to the present invention.

FIG. 5 is a configuration diagram of a fifth encoder of the present invention.

FIG. 6 is a configuration diagram of a sixth encoder of the present invention.

FIG. 7 is a configuration diagram of a seventh decoder of the present invention.

FIG. 8 is a configuration diagram of an eighth decoder of the present invention.

FIG. 9 is a flowchart of an algorithm used in a scanning pattern determination circuit.

FIG. 10 is a configuration diagram of a scanning pattern estimation circuit.

FIG. 11 is an example of a scanning pattern for each layer.

FIG. 12 is an explanatory diagram of a band division method using wavelet transform.

FIG. 13 is an explanatory diagram of a communication band structure of a filter bank.

FIG. 14 is a configuration diagram of a conventional encoder.

FIG. 15 is an explanatory diagram of a conventional scanning pattern.

FIG. 16 is a configuration diagram of a conventional decoder.

[Explanation of symbols]

 1 to 16 Two-dimensional band-pass filter 17 to 32 Sub-sampling circuit 103 Scan converter 104 Scan pattern determination circuit 105 Encoder 204 Scan pattern determination circuit 205 Encoder 206 Specific pattern replacement circuit 303 Inverse scan converter 304 Decoder 404 Decoder 405 Scan abort code replacement circuit 503 Scan converter 504 Scan pattern estimation circuit 505 Encoder 604 Scan pattern estimation circuit 605 Encoder 606 Specific pattern replacement circuit 703 Reverse scan converter 704 Decoder 705 Scan pattern estimation circuit 804 Decoding 805 Scan abort code replacement circuit 806 Scan pattern estimation circuit 2000 Layer separation circuit 2001 L-1 scan pattern determination circuit 2002 L-2 scan pattern determination circuit 2003 L-3 scan pattern Decision circuit 2004 Second scan pattern decision circuit 2005 First scan pattern decision circuit

Claims (8)

[Claims] In a sub-band coding method or a wavelet transform coding method in which an input signal is divided into frequency bands and coded, a transform coefficient within a grouped range is scanned and coded for each frequency band. Means for determining one or more scanning methods for each frequency band, and determining the scanning method so as to reduce the final code amount within the grouped range. A signal, comprising: means for scanning a transform coefficient for each frequency band according to the determined scanning method to form a transform coefficient sequence; and means for encoding the determined scanning method together with the transform coefficient sequence. Encoding method. 2. A signal encoding method according to claim 1, wherein, when a specific value of the transform coefficient sequence continues to the end of the grouped range, the continuous coefficient is replaced with a scan abort code. A signal encoding method in which, when determining a scanning method, the criterion for determining the scanning method is that the total amount of coefficients replaced by the scanning abort code is maximum. 3. A means for decoding code data output by the signal coding method according to claim 1 and separating the code data into a transform coefficient sequence and a scanning method, and said decoding in accordance with said decoded scanning method. A signal decoding method characterized by reconstructing transform coefficients from a transformed transform coefficient sequence and obtaining a group of coefficients capable of subband synthesis or inverse wavelet transform. 4. A means for decoding code data output by the signal coding method according to claim 2 and separating the code data into a transform coefficient sequence including a scan truncation code and a scanning method; Means for replacing the scan abort code in the transform coefficient sequence including the scan abort code with a continuous specific coefficient sequence, and reconstructing transform coefficients from the replaced transform coefficient sequence according to the decoded scanning method, subband synthesis or A signal decoding method characterized by obtaining a group of coefficients capable of inverse wavelet transform. 5. In a sub-band coding system or a wavelet transform coding system in which an input signal is divided into frequency bands and coded, quantized transform coefficients within a grouped range are converted from coefficients in a low frequency band. A method of sequentially scanning and encoding coefficients in a high frequency band, wherein one or more scanning methods can be used for each frequency band, and quantization of a low frequency band within the grouped range. Means for determining a scanning method in a higher frequency band from the converted conversion coefficients, means for scanning a conversion coefficient for each frequency band according to the determined scanning method to form a conversion coefficient sequence, and A signal encoding method comprising encoding means. 6. A signal encoding method according to claim 5, wherein, when a specific value in the transform coefficient sequence continues to the end of the grouped range, the continuous coefficient is replaced with a scan abort code. A signal encoding method characterized by further comprising means. 7. A means for decoding code data output by the signal encoding method according to claim 5 into a transform coefficient sequence, and decoding belonging to a low frequency band in the restored transform coefficient sequence. Means for determining a scanning method in a higher frequency band from the already-completed conversion coefficients by the same algorithm as the encoder according to claim 5, and transform coefficients from the decoded conversion coefficient sequence according to the determined scanning method. A signal decoding method characterized by sequentially restoring from a low-frequency band to a high-frequency band to obtain a group of coefficients capable of subband synthesis or inverse wavelet transform. 8. A means for decoding code data output by the signal encoding method according to claim 6 into a transform coefficient sequence including a scan truncation code, and a transform including the decoded scan truncation code. Means for converting the scanning truncation code in the coefficient sequence into a continuous specific coefficient sequence to obtain a decoded transform coefficient sequence; and a higher frequency band from the decoded transform coefficient belonging to the low frequency band in the restored transform coefficient sequence. 7. A means for determining a scanning method by the same algorithm as that of the encoder according to claim 6, and a method of restoring transform coefficients from the decoded transform coefficient sequence according to the decoded scanning method, and subband synthesis or inverse wave. A signal decoding method characterized by obtaining a group of coefficients that can be transformed.