JP3114796B2 - Variable length coding apparatus and method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method for coding an image signal by dividing the image signal into a plurality of frequency bands, and more particularly to compression of an information amount.

[0002]

2. Description of the Related Art Generally, an image signal has an enormous amount of information, and various attempts have been made to reduce the amount of information by compressing it when recording or transmitting. As a method of compressing image data, a sub-band encoding method is known in which an image signal is divided into a plurality of frequency bands (hereinafter, referred to as sub-bands) using a filter and compressed.

The operation flow of general image encoding / decoding will be described with reference to FIG. The encoding step converts the input image into frequency components 2001
(For example, Wavelet transform), a step 2002 for quantizing the result, and a step 2004 for performing variable length encoding on the result of the quantization. Data compression is realized by the following method.

(A) An energy distribution is biased by converting an input image into frequency components. (b) By quantizing the input image changed to the frequency component, weighting is performed for each frequency component, and the occurrence frequency of data is biased. (c) By performing variable-length coding on the quantization result, a short code is assigned to high-frequency data, a long code is assigned to low-frequency data, and compression can be performed by shortening the average code length as a whole.

[0005] In the quantization step 2002, a quantization table 2003 for designating bit allocation for each frequency component is provided.
Is used, and in the variable length encoding step 2004, the Huffman table 20 based on the frequency of occurrence (entropy) is used.
05 is used.

On the other hand, the decoding step comprises a variable length decoding step 2007, an inverse quantization step 2009, and an inverse transformation step 2011.

Next, subband coding will be described with reference to FIG. In FIG. 21, reference numeral 2101 denotes a horizontal high-pass filter (HPF), reference numeral 2102 denotes a horizontal low-pass filter (LPF), reference numerals 2103 and 2104 denote horizontal downsamplers for performing ダ ウ ン downsampling in the horizontal direction, 2105 and 2107. Is the vertical HPF, 210
6, 2108 are vertical LPFs, 2109, 211
Reference numerals 0, 2111, and 2112 denote vertical downsamplers that perform ダ ウ ン downsampling in the vertical direction.

FIG. 22 shows an example of division in the case where an image signal is divided into bands by the configuration shown in FIG. FIG. 22A shows a state in which the image signal is equally divided into two bands in both the horizontal direction and the vertical direction, and is divided into a total of four subbands. The names of the sub-bands indicate the states of the sub-bands in the horizontal and vertical directions in order, where L indicates a low band and H indicates a high band.

The wavelet transform coding method is a type of subband coding, in which a low frequency band (LL component in FIG. 22A) is finely divided, and hierarchical subband division is performed.

FIG. 22 (b) shows an example of division by the wavelet transform method. In the figure, image data is divided into a total of ten sub-bands having different bandwidths.
The number added to each subband indicates the subband number. Hereinafter, a case will be considered in which data divided into these ten subbands is scanned, quantized, and subjected to variable length coding.

Conventional scanning and variable length coding methods include the following two methods. The first method (method A is a method in which scanning is performed independently for each band to perform quantization and variable-length coding. In this case, one-dimensional Huffman coding is used as the variable-length coding method. This is common, for example, "IEEE Transactions on Communications,
Scene Adaptive Coder by Wen-Hsiung Chen and Willi
amoK. Pratt, Vol. Com-32, No. 3, March 1984, P
. 225-232 "shows the details.

In the above-mentioned paper, a table of two Huffman codes as shown in FIG. 23 is used to encode coefficient data after quantization. According to the table of FIG.
Encode non-zero quantized coefficients. At that time, only the absolute value of the coefficient is encoded. This is because coefficients having the same absolute value appear at the same frequency. One bit is transmitted following the Huffman code for the sign. One code word is assigned to one absolute value in 12 entries.
Other values (absolute values) are transmitted after adding the coefficient data as they are following the “OTHER code”.

From the table shown in FIG. 23B, run-length encoding is performed on the zero coefficient. Therefore, in order to distinguish the two tables at the time of decoding, the run length prefix (RL PREF) is used as the identification code of the run length code.
IX) is used.

The tables shown in FIGS. 23A and 23B are created based on the data value (absolute value) to be subjected to variable length coding or the occurrence frequency of the run length length. In FIG. 23A, the absolute value = 値 1 ′, and the occurrence frequency is extremely high, so that a 1-bit code is assigned. On the other hand, values such as absolute value = $ 9 '* to $ 12' * are 8-bit codes because their occurrence frequency is not so high. The variable-length code compresses the data amount by expressing such a frequent coefficient value with a short code length.

In addition, the present method is called one-dimensional Huffman coding because coding is performed by separately assigning a Huffman code to a coefficient other than zero and a zero run.

Next, a second method (method B) is a method in which subband frequency components are scanned so that the coding result becomes 8 × 8 block coefficients, and quantization and variable length coding are performed. In this case, as a variable-length coding method,
It is assumed that two-dimensional Huffman coding described below will be performed.

As a scanning method for obtaining a block coefficient, there is a "coding method for hierarchically expressed signals" as described in Japanese Patent Application Laid-Open No. 4-245683. this is,
As shown in FIG. 24, the object is to scan from low frequency to high frequency out of the 10 frequency components after the Wavelet transform to obtain a correlated 8 × 8 block coefficient.

In the two-dimensional Huffman coding, a Huffman code is assigned as a set of coefficient data other than zero and zero run.
There is a “bit rate reduction method and device” as described in No. 0. This method is employed in JPEG and MPEG because it is more efficient in encoding coefficient data obtained by dividing a block such as 8 × 8.

The present inventors use a general TV broadcast video scene and perform a Wavelet conversion step 20 in FIG.
01, the same processing was performed up to the quantization step 2002, and the above two methods were compared as the variable length coding step 2004. The evaluation conditions are as shown in Table 1 below. As the quantization table, a table shown in Fig. 25 which is divided into a luminance (Y) component and a chrominance ((UV) component) is used.
The coefficient data after the velocity conversion is divided by the value shown in FIG. The filters (HPF, LPF) for performing the Wavelet transform shown in FIG. 21 include SSKF (S
ymmetyric Short Kernel Filter) was used.

[0020]

[Table 1]

Table 2 below shows the evaluation results of the coding efficiency. The comparison was made assuming that the data size after encoding in the method B was 100%.

[0022]

[Table 2]

As described above, when an image equivalent to the SIF size is compressed to about 1/10 by the wavelet transform, a result is obtained in which the one-dimensional Huffman method of independently encoding each band has higher encoding efficiency. Was. This is
It is considered that, because the quantization step of the high-frequency component was increased with respect to the compression ratio of the data to be evaluated, zero run, that is, many long continuous coefficients of zero occurred. That is, it is more efficient to encode a large block for each band than to encode a small 8 × 8 block.

Here, an apparatus for realizing the one-dimensional Huffman coding described in the above-mentioned paper is shown in FIG.
The detailed operation will be described with reference to FIG. FIG.
FIG. 6 is a block diagram showing an example of a conventional one-dimensional Huffman coding apparatus. For example, assuming that the image size in FIG.
X26, which is composed of 1040 coefficient data. Hereinafter, the operation of the zero-run encoding will be described first.

When the sub-band 3 is scanned, it is assumed that coefficients are formed in the following order. Coefficient [1], Zero run [3], Coefficient [2], Zero run [15], Coefficient [3], Zero run [503], Coefficient [1], Zero run [517] The zero determination unit 2601 stores one coefficient data. Loading,
It is determined whether or not the read coefficient data is 0.
If the coefficient data is not 0, the coefficient data is sent to the coefficient table decoding unit 2606, and the coefficient table 2605 is sent.
Is used to encode the coefficient data. The operation of encoding the coefficient data will be described later.

On the other hand, if the read coefficient data is determined to be 0 by the zero determination section 2601, the zero counter 26
03 is a continuous length of zero (hereinafter referred to as zero run length)
Is counted by 1. The zero counter 2603 is used until the zero determination unit 2601 determines that the coefficient data is not zero.
Continue this operation.

When the zero determination unit 2601 determines that the coefficient data is not zero, the zero counter 2603 sends the counted zero run length to the zero run encoding unit 2604, and clears the zero run length to zero.

The zero-run encoding unit 2604 encodes the transmitted zero-run length using the code table 2605. The code table for zero run is as shown in FIG. However, actually, the “RL” in FIG. 23A, which means a zero-run code at the beginning of the Huffman code in FIG.
The configuration is such that a “PREFIX” code is added.

FIG. 27 is a view for explaining the processing of the zero-run code section 2604. In the figure, each step performs the following processing. (2701) Zero run length zero is zero run length r (r>
0, r are integers and represent the maximum run length that can be processed). The case where the zero run length zero is larger than the zero run length r is shown below (270).
2) and (2703) are performed, and if smaller, the process of (2704) shown below is performed.

(2702) Encoding of the zero run length r is performed. (2703) The zero run length r is subtracted from the zero run length zero, and the process of (2701) is performed. (2704) The zero run length zero is encoded.

The above operation is performed until the coefficient data of subband 3 is exhausted.

The table size of the zero run is set to 255 (r =
FIG. 6 (a) shows the code result in the case of (255). Since the zero-run 503 has no code, 503/255 = 2, which is much more than 3, so it is represented by two codes of the zero-run 255 and one code of the zero-run 3.

When the last coefficient data is zero, the zero-run encoding unit 2604 does not encode the zero-run length at that time, but performs encoding using an EOB (End Of Block) code.

Next, the encoding operation of non-zero (coefficient) data will be described. Consider a case in which the coefficients of subband 3 are configured in the following order. Coefficient [1], coefficient [20], coefficient [18], coefficient [4
0], zero run [10], coefficient [-5], zero run [1]
[025] The zero determination unit 2601 reads the coefficient data one by one,
If the read coefficient data is determined not to be 0 by the zero determination unit 2601, the coefficient code generation unit 2607 performs encoding using the coefficient table 2605.

At the time of encoding, instead of independently assigning a code word to all coefficient values, as shown in FIG. 23 (a), a code word is independently assigned according to the frequency of occurrence. -12) processing and other "coefficient other code (AMPLITUDE1
3) The processing is performed in a format separated from the processing.

In operation, the coefficient table decoding unit 2606 compares the absolute value of the input coefficient data with the coefficient value (AMPLITUDE) of the coefficient table 2605 to determine a table number.

In the case of the above data string, the coefficient table decoding section 2606 outputs a table number “1” to the coefficient code generating section 2607 for the first coefficient value [1]. The coefficient code generator 2607 includes a code table 260
2 is read out using FIG.
Data selector 2609 in coefficient code format (a)
Output to If the table number is other than “1” to “12”, the data selection unit 2609 selects data from the coefficient other code generation unit 2608.

In the case of the second coefficient value [20], the coefficient table decoding unit 2606 does not match the coefficient value held in the coefficient table 2605.
3 "is output to the coefficient OTHER code generation unit 2608. The coefficient OTHER code generation unit 2608 reads the assigned code word using the code table 2602, and uses the coefficient OTHER code format shown in FIG. At this time, if the number of effective bits of the original coefficient data is 12 bits, the coefficient value of 12 bits is combined with the OTHER code and output.

The above operation is performed until the coefficient data of subband 3 is exhausted. FIG. 12A shows the output result of the variable length coding device at this time.

[0040]

The conventional variable-length coding apparatus and method are configured as described above. By dividing an image signal into a plurality of frequency bands and performing coding, an enormous amount of information can be obtained. Can be recorded or transmitted by reducing the amount of information by compressing an image signal having the following. However, it has the following three problems.

(1) As shown in FIG. 23 (b), the table size for the zero-run length code is as large as 30 entries, and the circuit scale is large. (2) A long zero run must be represented by repeating the code of the zero run 255, and the longer the zero run length, the more redundant the code result. This evaluation image (3
20 × 208), the number of coefficient data of the largest subband (subbands 8, 9, and 10) is 160 × 100 = 1.
6000, which is considered to be a very long zero run, further increasing the redundancy and deteriorating the coding efficiency. (3) For encoding non-zero coefficients, the coefficient OTHE
When generating the R code, the number of bits of the original coefficient value is directly used as the encoding target data. However, the number of effective bits of the coefficient data has become smaller than the number of bits of the original coefficient value due to the quantization processing. Therefore, the coding efficiency is reduced in the conventional variable-length coding apparatus that handles coefficient data with a fixed length.

The present invention has been made in view of the above problems, and a variable length code capable of suppressing the redundancy of zero run length encoding without preparing a large size zero run code table. It is an object of the present invention to obtain a chemical conversion device and method.

It is another object of the present invention to provide a variable length coding apparatus and method capable of suppressing the redundancy of non-zero coefficient coding according to the number of effective bits after the quantization processing.

[0044]

The first variable length coding method and apparatus according to the present invention include a group table and a zero-run grouping means. After grouping according to the size, this is encoded, so that even with a long zero run length, the zero run length can be efficiently compressed with a small zero run table size.

The second variable length coding method and apparatus according to the present invention include coefficient coding means corresponding to an arbitrary number of coefficient bits, and when coefficient data is not zero (non-zero), the coefficient data is coded and quantized. Since the encoding is performed together with the number of effective bits of the quantized coefficient data, coefficient encoding according to the number of effective bits after quantization can be performed, and the encoding efficiency of non-zero coefficient data can be improved. It becomes possible.

According to the third variable length coding method and apparatus of the present invention, coefficient values are grouped for each predetermined range and coding is performed. Encoding efficiency can be increased even in encoding of increasing low frequency components.

The fourth variable-length coding method and apparatus according to the present invention include means for selecting a grouping method of coefficient data, and optimizes a method of assigning a code word to a coefficient value for each subband. Therefore, encoding can be performed more efficiently.

[0048]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1 FIG. FIG. 1 shows a block diagram of an encoding device used in a variable length encoding device and method according to Embodiment 1 of the present invention. In FIG. 1, 101 is a zero determination unit, 102 is a coefficient sign unit, 103 is a zero counter, 10
4 is a zero-run group decoder, 105 is a zero-run encoder, 106 is a group table, and 107 is a code table.

Next, the operation will be described. Here, it is assumed that the image data is divided into ten sub-bands by sub-band coding as shown in FIG. Each sub-band is composed of quantized coefficient data. The number described in the subband indicates the subband number. For example, if the image size in FIG.
Assuming 20 × 208, the size of subband 3 is 40 ×
26, which is composed of 1040 coefficient data. Each subband is independently encoded.

When the sub-band 3 is scanned, it is assumed that the sub-band 3 is configured in the following order. Coefficient [1], Zero run [3], Coefficient [2], Zero run [15], Coefficient [3], Zero run [503], Coefficient [1], Zero run [517] Coefficient data constituting the subband is read out in order. And sent to the zero determination unit 101. The zero determination unit 101 determines whether the transmitted coefficient data is 0.

When the zero determination section 101 determines that the value is 0, the zero run counter 103 counts the zero run length by one.

The zero counter 103 continues counting until the zero determination unit 101 determines that the coefficient data is not 0.

In the zero determination section 101, the coefficient data is 0
Otherwise, the zero-run counter sends the current zero-run length to the zero-run group decoding unit 104, and
Clear the zero run length.

The zero-run group decoding unit 104 determines a group number using the group table 106 based on the transmitted zero-run length.

FIG. 2 shows the group table 106. As shown in FIG. 2, the zero run length is divided into 16 groups. In other words, the zero run length 1 that appears frequently
7 are independent groups as they are, and the other zero run lengths are grouped in a range that can be expressed by a power of 2.

As for the zero run length 2048 or more, it is considered that the frequency of occurrence is considered to be extremely low. It is considered that the higher the sub-band, the more often the sub-band ends with zero run, and the last zero run is omitted in EOB (End Of Block).

The number of effective bits indicates the number of bits effective to represent the zero run length. FIG. 3 shows the zero run length in each group. The zero run length is composed of 15 bits, and LSB (least significant bit; Least
Siginificant Bit). The number of bits effective to represent the zero run length is (the minimum number of bits required to represent the zero run length) -1. This is because the zero-run groups are grouped in a range that can be expressed by a power of two.

Next, the zero run group decoding section 104
Is the MSB of the zero run length (most significant bit; Most Signifi
bit from the cant Bit) side
The processing is interrupted at the time when the bit reaches 1 where a 1 is set, and the group number is determined from the bit length of consecutive zeros.

This processing is based on the fact that the zero-run groups are grouped in a range that can be expressed by a power of two. For example, if the zero run length is included in the group 10, 1 is always set to the sixth bit from the LSB side, and if the zero run length is included in the group 13, the LSB is LSB.
1 always stands at the ninth bit from the side.

For example, the zero run length 8-15 included in the group 8 can be expressed with a minimum of 4 bits, and it is clear that 1 is always set to the most significant bit.
This makes it possible to omit the most significant bit. In this case, the number of effective bits is three. However, for group 16, the number of effective bits is calculated based on the maximum zero run length 32767.

FIG. 4 shows a zero run group decoding unit 104.
It is a flowchart figure explaining the said process. That is, in step (401), a variable (bit) that counts 0 bits is cleared. Next, in step (402), the bit position is set to the MSB side. At step (403), it is determined whether or not the current bit is 1, and if it is 0, the process proceeds to step (404) and bi is applied.
t is incremented by 1, and the same process is repeated until the current bit is determined to be 1 in step (403). If it is determined in step (403) that the current bit is 1, the process proceeds to step (405), where bit
The group number is determined from the value (size). Then
Proceeding to step (406), the group number is sent to the zero-run encoder 105, and the process ends.

The zero-run encoding unit 105 encodes the zero-run length using the group number sent from the zero-run group decoding unit 104, the zero-run length, and the number of effective bits of the zero-run length obtained from the group table 106. Do.

FIG. 5 shows the coded zero run length of each group. The zero run length is encoded by storing the zero run length represented by the number of effective bits after the code of the group from the LSB.

The coefficient data determined by the zero determination section 101 to be not 0 is sent to the coefficient coding section 102, encoded using the code table 107, and transmitted as code data. Then, the above processing is performed until the coefficient data of subband 3 is exhausted. This encoding result is shown in FIG.
Shown in From the group table 106 shown in FIG. 2, the zero run 15 and the zero run 503 belong to the group 8 and the group 13, respectively. Since the effective bit numbers are 3 bits and 8 bits respectively, FIG. The encoding result is as shown.

With the above processing, encoding of subband 3 has been realized. By performing this processing on all of the subbands 1 to 10, compressed data can be obtained.

Table 3 below shows the evaluation results of the coding efficiency described in the conventional description. The evaluation conditions are the same as those of the conventional description. As the method of the present embodiment, the same processing as the conventional method is performed for the encoding of the non-zero coefficient, and the grouping shown in FIG. .

[0067]

[Table 3]

As shown in the above table, when the conventional method B is set to 100%, the result is almost the same as the method A or the efficiency is slightly improved. Here, the conventional method A
Considering that this requires 30 entries of the zero run table, it can be seen that this embodiment achieves the same coding efficiency with about half the entries. Therefore, it is an extremely effective method to group and encode zero runs as shown in FIG.

In this embodiment, the zero run length is set to 1
Although the data is divided into six groups, the grouping method may be changed, for example, by increasing or decreasing the zero run length to which codewords are directly assigned.

As described above, according to the present embodiment, it is determined whether or not the quantized coefficient data is zero. If it is determined that the coefficient data is not zero, the coefficient data is converted into a code table using the code table. When the coefficient data is determined to be zero, the number of consecutive zeros is counted until the coefficient data is determined to be not zero, and a zero run length is obtained. Since the data is classified into a plurality of groups according to the frequency of appearance and is encoded in units of the group, even a long zero run length can be grouped and encoded, and the zero run length can be efficiently compressed with a small zero run table size. Will be able to

Embodiment 2 Next, a variable length coding apparatus and method according to the second embodiment will be described. FIG. 7 is a block diagram of a variable length coding device used in the variable length coding device and method according to the second embodiment. In FIG. 7, reference numeral 701 denotes a quantization unit for quantizing coefficient data;
Is a zero determination unit that determines whether zero is included in the quantized coefficient data, 703 is a zero run counter that counts zero runs, 704 is a code table, 705 is a zero run encoding unit, 706 is a coefficient table, For example, it manages the information shown in FIG. 707 is a coefficient table decoder, 708 is a coefficient code generator, 709 is a coefficient OTHER code generator, and 710 is a data selector.

Next, the operation will be described. Here, the image data is sub-band coded as shown in FIG.
Consider the case of dividing into ten subbands. The number described in the subband indicates the subband number. For example, if the image size in FIG. 22B is 320 × 208, the size of subband 3 is 40 × 26 and 1
This means that the data is composed of 040 coefficient data.

Each sub-band is independently quantized by the quantization unit 7.
Quantization is performed by 01 and variable length coding is performed. Here, the quantization unit 701 converts the 12-bit coefficient data into, for example, the data shown in FIG.
Assuming that the division is performed by the quantization step shown in (a), the number of effective bits of the coefficient data of subband 3 is 10 bits instead of 12 bits.

Now, consider a case where the quantized subband 3 is scanned and coefficients are formed in the following order. Coefficient [1], coefficient [20], coefficient [18], coefficient [4
0], zero run [10], coefficient [-5], zero run [1]
[025] The zero determination unit 702 reads coefficient data one by one, and determines whether the read coefficient data is zero. If it is determined that the coefficient data is 0, the zero-run counter 703 counts the length of the continuous zero, and the zero-run encoding unit 705 encodes the zero-run length using the code table 704.

On the other hand, when the read coefficient data is judged not to be 0 by the zero judgment section 702, the coefficient table decoding section 707 encodes using the coefficient table 706. As the format of the encoded data, FIG.
This is performed by using a "coefficient code" for assigning only codewords according to the frequency of occurrence, and a "coefficient OTHER code" for connecting coefficient data to codewords shown in FIG. This is realized using the coefficient table as shown in FIG. 8 used in the first embodiment. FIG. 9 is a block diagram for explaining the operation of the coefficient table decoding unit 707. FIG.
901 is a comparator that performs coefficient table comparison, stores each coefficient value to which a code word is independently assigned in FIG. 8 in a register, compares this with input data, and obtains a comparison result. If they match, "1" is output. The address conversion unit 902 decodes the table number based on the output of the comparator 901. The decoding result is an address as shown in the coefficient Huffman table 904, for example.

In the coefficient Huffman table 904, when the output result is all 0, it is determined that the code is a code other than 1 to 15, and the code is made to correspond to the table number 16. Reference numeral 903 denotes a coefficient Huffman table in the code table 704.

FIG. 10 shows a coefficient table decoding unit 70.
FIG. 7 is a flowchart illustrating a coefficient code generation process in a coefficient code generation unit 708, a coefficient other code generation unit 709, and a data selection unit 710. In each step, the (1001) coefficient table decoding unit 707 takes the absolute value of the input coefficient data. (1002) The coefficient table decoding unit 707 decodes the table number by the processing described above, and
If it is 6, the process of step (1006) is performed, and if it is not 16, the process of step (1003) is performed. (1003) A Huffman code uniquely assigned to the table number is obtained from the code table 704 (this is
code). (1004) The code bit is shifted by the hcode length (this is referred to as scode). (1005) Code data is generated by the logical sum of the hcode and the scode. This data A is shown in FIG.
The data has the format shown in FIG. (1006) The Huffman code assigned to the table number 16 is obtained from the code table 704 (this is
r). (1007) Coefficient data (including sign bit) is
Shift by r lengths (this is ccode). At this time, the coefficient data uses only the number of effective bits after quantization. (1008) Code data is generated by the logical sum of other and ccode. This data B has the format shown in FIG. Thus, the above-described data sequence is as follows.

First, the coefficient table decoding unit 707
The table number “1” is output to the coefficient code generator 708 for the first coefficient value [1]. Coefficient code generator 708
Reads the assigned code word using the code table 704 and outputs the code word to the data selection unit 710 in the coefficient code format shown in FIG. Data selector 71
If the table number is other than "16" in step (1002), 0 outputs the data from the coefficient code generator 708 as a variable length code through steps (1003) to (1005). In the second coefficient value [20], in step (1002), the coefficient table decoding unit 707 outputs the table number “16” to the coefficient OTHER code generation unit 709 because all the coefficient values in the coefficient table 706 do not match. . Coefficient OT
The HER code generation unit 709 reads out the assigned code word using the code table 704, and
006) through step (1008), the data selection unit 710 in the coefficient OTHER code format shown in FIG.
Output to

At this time, the coefficient OTHER code generator 70
Numeral 9 obtains the information of the quantization step shown in FIG. 25A from the quantization unit 701 and combines only the coefficient value of the number of effective bits after quantization with an OTHER code (a 16th code) and outputs it. I do. That is, the coefficient OTHER code generation unit 709
Means that any length of 12 bits or less can be selected as coefficient data following the code word.
Using the lower 10 bits, the coefficient O as shown in FIG.
Synthesize with THER code.

The above operation is performed until the coefficient data of subband 3 is exhausted. FIG. 12B shows the output result of the variable length coding device at this time. As compared with the coding result (FIG. 12 (a)) of the conventional device, the coefficient OTHER code, that is, the code data portion by the coefficient value of the number of effective bits after quantization is reduced.

The encoded data can be decoded by managing the variable length code data and the quantization table shown in FIG. 25 together.

Table 4 shows the evaluation results of the coding efficiency described in the conventional description and the comparison results according to the second embodiment. The evaluation conditions are the same as those shown in Table 1. As the method of the second embodiment, the comparison is made assuming that the zero-run coding uses the same method as the conventional method.

[0083]

[Table 4]

Assuming that the data size at the time of encoding by the conventional method B is 100%, the result is that the data size can be further reduced by about 8% compared with the method A. In particular, when the circuit size is taken into consideration and the sub-band division is a power of 2, as in the quantization table shown in FIG.

In the coefficient table of FIG. 8, the 12-bit coefficient data is divided into 16 pieces. However, the total number may be changed by, for example, increasing or decreasing the coefficient value to which a codeword is directly assigned.

Also, in FIG. 8, coefficient values 1 to 15 are assigned to the coefficient absolute values of the table numbers 1 to 15, respectively. May be assigned. Also, the coefficient table size is not particularly limited to 16 entries, but may be any other size.

If the size of the coefficient table becomes large, FIG.
As shown in (1), the number of coefficient value registers increases in proportion thereto, and the circuit scale including the decoding circuit 901 increases.

Also, in the case of processing by software, it is necessary to read the coefficient values and to compare the read coefficient values until the input coefficient data matches the coefficient values. , Number of processes,
The processing time increases together.

FIG. 13 shows the relationship between the table size and the code amount. Each polygonal line represents the code amount for the table size in each subband, and the number at the left end indicates the subband number. The evaluation conditions are the same as the evaluation in the second embodiment.

The code table 704 determines the frequency of occurrence of coefficient data constituting a subband for each subband, and changes the table size between 2 and 20.

As shown in FIG. 13, the code amount decreases as the table size increases. This is because, when the table size is increased, the coefficient value to which the Huffman code can be independently assigned increases, and the encoding efficiency increases.

However, the graph in each subband follows a gradual change from a table size of about 8 and hardly changes from a table size of about 16. When the table size is 20 or more, it is omitted because there is almost no difference from the case where the table size is 20.

It is considered that this is because the variation in coefficient data existing in each subband is small due to the quantization in the quantization section 701. That is, the ratio of coefficient data serving as an OTHER coefficient is small in coefficient data constituting a subband.

According to the above result, according to the second embodiment, a coefficient encoding means corresponding to an arbitrary number of coefficient bits is provided, and when coefficient data is not zero (non-zero), it is encoded and quantized. Since the encoding is performed together with the number of effective bits of the coefficient data afterward, coefficient encoding according to the number of effective bits after quantization can be performed. In particular, by setting the table size to 8 to 16, The encoding efficiency of zero coefficient data can be increased, and the most optimal encoding can be realized in each aspect of the compression ratio, the circuit scale, and the processing time.

Embodiment 3 Next, a variable length coding apparatus and method according to the third embodiment will be described. FIG. 14 is a block diagram of a variable length coding device used in the variable length coding device and method according to the third embodiment. 14, components having the same functions as those of the encoding device in FIG. 7 are denoted by the same reference numerals, and description thereof will be omitted. In the figure,
Reference numeral 1401 denotes a coefficient group table, for example, which manages information shown in FIG. 15, 1402 denotes a coefficient group decoding unit, and 1403 denotes a corresponding coefficient OTHER code generation unit.

The operation is almost the same as that described in the second embodiment, but in this embodiment, the coefficient O
The method for generating the THER code is different. Similar to the second embodiment, the coefficient group table 1401 shown in FIG. 14 considers a case where image data is divided into ten subbands by subband encoding as shown in FIG.
The numbers described in FIG. 22B indicate the subband numbers.

Each sub-band is independently quantized by the quantization unit 7.
At 01, quantization and variable length coding are performed based on FIG. For example, when the sub-band 3 is scanned, it is assumed that the coefficients are arranged in the following order. Coefficient [1], coefficient [20], coefficient [18], coefficient [4
0], zero run [10], coefficient [-5], zero run [1]
[025] The zero determination unit 702 reads coefficient data one by one, and determines whether the read coefficient data is zero. If the coefficient data is 0, zero-run counter 703 obtains information of a length in which zero continues, and a zero-run encoder 705
Encodes the zero run length using the code table 704.

On the other hand, when the read coefficient data is determined not to be 0 by the zero determination section 702, the coefficient group decoding section 1402 determines a group number from the transmitted coefficient data using the group table 1401.

Here, the coefficient group table 1401
15 is shown in FIG. As shown in FIG. 15, the coefficient values are divided into 15 groups. The coefficient absolute values 1 to 7 having a high appearance frequency are regarded as independent groups as they are, and the other coefficient absolute values are divided within a range that can be expressed by a power of 2, and 8
The number of effective bits is associated with the third and higher groups. The number of effective bits is (the minimum number of bits required to represent the coefficient absolute value) −1. This is because the coefficient absolute values are grouped in a range that can be expressed by a power of two.

FIG. 16 shows coefficient data corresponding to each group. The 12-bit coefficient data is assumed that the effective bits are stored in LSB justified by quantization, and the absolute value is 11 bits at the maximum.
For example, the coefficient absolute value included in group 8 is 8-15
And can be represented with a minimum of 4 bits, and always 4
It is clear that 1 is placed in the bit. This makes it possible to omit four bits, and in this case, the number of effective bits is three.

The coefficient group decoding section 1402 determines bits one bit at a time from the MSB side of the coefficient data determined to be non-zero, and suspends the processing when it reaches the bit where 1 is set. The group number is determined from the position.

This processing is based on the fact that the coefficient values are grouped in a range in which the coefficient values can be represented by a power of two. For example, for a coefficient value included in the group 10, 1 is always set to the sixth bit from the LSB side, and for a coefficient value included in the group 13, 1 is always set to the ninth bit from the LSB side.

FIG. 17 shows a coefficient group decoding section 1402.
It is a flowchart figure explaining the said process. In each step, the following processing is performed. (1701) Variable (bi for counting bits that are 0)
Clear t). (1702) The bit position is set on the MSB side. (1703) Determine whether the current bit is 1 and
If it is 0, the process of step (1704) is performed, and 1
If it is, the process of step (1705) is performed. (1704) Increment the bit by 1, move the comparison position to the right by one bit, and return to step (1703). (1705) The group number is determined from the bit value. (1706) The group number is sent to the coefficient code generation unit 708 and the coefficient OTHER code generation unit 1403. In accordance with the group number, the coefficient code generator 708 and the coefficient other code generator 1403 generate codes as shown in FIG.

[0104] The coefficient OTHER code generation unit 1403
Code table 704 different code words depending on the group number
Then, the coefficient data corresponding to the number of effective bits in the coefficient group table of FIG.

When the above-described coefficient sequence is input, coefficient group decoding section 1402 outputs group number “1” to coefficient code generation section 708 from the first coefficient value [1]. The coefficient code generation unit 708 reads the allocated codewords using the code table 704, and outputs the read codewords to the data selection unit 710 in the format of Group 1 in FIG. The data selection unit 710 determines that the table number is “1” to
If “7”, the data from the coefficient code generation unit 708 is output as a variable length code.

For the second coefficient value [20], coefficient group decoding section 1402 sets group number “8” to coefficient O
Output to the THER code generation unit 1403. Coefficient OTH
The ER code generation unit 1403 reads out the assigned code word using the code table 704, extracts the number of effective bits of the group number “8” from the coefficient group table 1401, and obtains the “group 8 code word” as shown in FIG. + Sign bit + coefficient absolute value (3 bits) ".

The above operation is performed until the coefficient data of subband 3 is exhausted. The output result of the variable length coding device at this time is shown in FIG. In the encoding result of FIG. 12C, the coefficient OT is smaller than that of FIG. 11B of the second embodiment.
It is considered that the length of the coefficient value of the HER code is optimized for each group, and the coding efficiency is improved.

By repeating the above processing for each of the divided subbands, variable length coding can be realized. In the coded data, the number of bits is associated with each group, and the decoding process can be performed without managing the data together with the quantization information shown in FIG.

In the present embodiment, the 12-bit coefficient data is divided into 15 groups.
5 may be merged into one group.

It is apparent that even if the number of groups and the range of grouping are changed in accordance with the compression ratio, it can be realized by the same encoding step as in the present embodiment.

As described above, according to the third embodiment, the encoding is performed by grouping the coefficient values for each predetermined range. In addition to the effects of the first and second embodiments, Also, the encoding efficiency can be improved even in the encoding of low frequency components in which the degree of dispersion of coefficient values increases.

Embodiment 4 Next, a variable length coding apparatus and method according to the fourth embodiment will be described. FIG. 19 is a block diagram of a variable length coding device used in the variable length coding device and method according to the third embodiment. In FIG. 19, components denoted by the same reference numerals as those in FIGS. 7 and 14 have the same functions, and a description thereof will be omitted. In FIG. 19, reference numeral 1901 denotes a table switching unit,
Reference numeral 1902 denotes a coefficient group decoding unit 14 according to the third embodiment.
Unlike the coefficient group 02, the coefficient group decoding unit includes a coefficient table 706 and a coefficient group table 1401.

The operation of the variable length coding apparatus is different in that the variable length coding apparatus has a coefficient table 706 and a coefficient group table 1401, and performs variable length coding by switching every subband.

Also, it is assumed that the coefficient is managed by adding the number of effective bits at the time of quantization to the 16th OTHER group of the coefficient table 706, and the coefficient OTHER code generation unit 1903 reads the effective bit number from the coefficient table 706. And outputs the code data. The operation after selecting the table is the same as that described in the second and third embodiments.

Generally, in a low-frequency subband (for example, 1 to 4 in FIG. 22B) in which the quantization step is reduced, the coefficient value variance increases and the ratio of the coefficient OTHER code increases. Table 1401 can be encoded more efficiently. Conversely, in the high frequency subbands (eg, 5 to 10 in FIG. 22B), the quantization step is increased, so that the variance of the coefficient values is biased.
No. 6 can be encoded more efficiently. Therefore, if these can be selected for each subband, coding can be performed very efficiently.

As described above, according to the fourth embodiment, the means for selecting a method of grouping coefficient data is provided, and the method of allocating code words to coefficient values for each subband is optimized. Coding can be performed efficiently.

[0117]

As described above, the variable length coding method and apparatus according to the present invention include the group table and the zero-run grouping means. After grouping according to the size, this is encoded, so that even with a long zero run length, the zero run length can be efficiently compressed with a small zero run table size.

Further, the variable length coding method and apparatus according to the present invention include a coefficient coding means corresponding to an arbitrary number of coefficient bits, and when the coefficient data is not zero (non-zero), the coefficient data is coded and quantized. Since the encoding is performed together with the number of effective bits of the quantized coefficient data, coefficient encoding according to the number of effective bits after quantization can be performed, and the encoding efficiency of non-zero coefficient data can be improved. There is an effect that can be.

In the variable length coding method and apparatus according to the present invention, the coefficient values are grouped for each predetermined range and coding is performed. Therefore, even if the zero run length is long, the zero run table size can be reduced. In addition to the above-described effect that the length can be efficiently compressed, there is an effect that the encoding efficiency can be increased even in encoding of a low-frequency component in which the variance of the coefficient value increases.

Further, the variable length coding method and apparatus according to the present invention include means for selecting a grouping method of coefficient data, and optimizes a method of assigning a code word to a coefficient value for each subband. Therefore, there is an effect that encoding can be performed more efficiently.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a variable-length encoding device that groups zero-run lengths according to Embodiment 1 of the present invention.

FIG. 2 is a diagram showing an example of a group table for zero-run encoding according to Embodiment 1 of the present invention.

FIG. 3 is a diagram showing zero run lengths grouped in FIG. 2;

FIG. 4 is a diagram illustrating processing of a zero-run grouping unit according to the first embodiment of the present invention.

FIG. 5 is a diagram illustrating a code data format having a zero run length according to the first embodiment of the present invention.

FIG. 6 is a diagram illustrating an example of an encoding result according to the first embodiment of the present invention and a conventional method.

FIG. 7 is a block diagram illustrating a configuration of a variable-length encoding device that performs non-zero coefficient encoding according to a quantization step according to Embodiment 2 of the present invention.

FIG. 8 is a diagram showing an example of a non-zero Huffman table according to Embodiment 2 of the present invention.

FIG. 9 is a block diagram showing a configuration of a coefficient table decoding unit according to Embodiment 2 of the present invention.

FIG. 10 is a diagram illustrating a process of a coefficient table decoding unit.

FIG. 11 is a diagram illustrating a non-zero code data format according to the second embodiment of the present invention.

FIG. 12 is a diagram illustrating an example of an encoding result according to Embodiments 2 and 3 of the present invention and a conventional method.

FIG. 13 is a diagram illustrating a relationship between a table size and a compressed code amount according to the second embodiment of the present invention.

FIG. 14 is a block diagram illustrating a configuration of a variable-length encoding device that performs non-zero coefficient grouping according to Embodiment 3 of the present invention.

FIG. 15 is a diagram showing an example of a group table for non-zero coefficient encoding according to Embodiment 3 of the present invention.

FIG. 16 is a diagram showing non-zero coefficient data grouped in FIG.

FIG. 17 is a diagram illustrating processing of a coefficient group decoding unit according to Embodiment 3 of the present invention.

FIG. 18 is a diagram illustrating a code data format of a non-zero coefficient subjected to grouping according to the third embodiment of the present invention.

FIG. 19 is a block diagram illustrating a configuration of a variable-length coding device including a plurality of non-zero coefficient coding methods according to Embodiment 4 of the present invention.

FIG. 20 is a basic block diagram of a subband encoding scheme.

FIG. 21 is a diagram showing a sub-band division method in the Wavelet transform method.

FIG. 22 is a diagram illustrating an example of division of frequency components in the Wavelet transform method.

FIG. 23 is a diagram showing an example of a Huffman table in a conventional variable-length coding scheme.

FIG. 24 is a diagram illustrating an example of a zigzag scan in a case where variable length coding is performed by dividing into 8 × 8 blocks in the Wavelet transform method.

FIG. 25 is a diagram illustrating an example of a quantization table in the Wavelet transform method.

FIG. 26 is a block diagram showing a configuration of a conventional variable length coding device.

FIG. 27 is a diagram for explaining the operation of the zero-run encoder in the block diagram of FIG. 25;

[Explanation of symbols]

101, 702 Zero judgment unit, 102 Coefficient sign unit, 1
03,703 Zero counter, 104 Zero-run group decoder, 105,705 Zero-run encoder, 1
06 group table, 107, 704 code table, 701 quantization unit, 706 coefficient table, 707
Coefficient table decoding unit, 708 coefficient code generation unit, 709, 1403, 1903 coefficient OTHER code generation unit, 710 data selection unit, 1401 coefficient group table, 1402, 1902 coefficient group decoding unit, 1901 table switching unit

──────────────────────────────────────────────────の Continuing on the front page (72) Inventor Yasuhiko Yamane 1006 Kadoma, Kazuma, Osaka Prefecture Inside Matsushita Electric Industrial Co., Ltd. (72) Inventor Takumi 1006 Kazuma, Kadoma, Osaka Pref. (56) References JP-A-63-132530 (JP, A) JP-A-63-212287 (JP, A) JP-A-63-263983 (JP, A) JP-A-6-350854 (JP, A) JP-A-6-237184 (JP, A) JP-A-6-152988 (JP, A) JP-A-6-77843 (JP, A) Japanese Translation of PCT Application No. 5-507598 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) H03M 7/46

Claims (14)

(57) [Claims] A variable input means for dividing an input signal into coefficient data representing a plurality of different frequency bands, performing quantization on the coefficient data for each of the frequency bands, and encoding the quantized coefficient data; A length encoding device, a zero determination unit that determines whether the coefficient data is zero, a code table that stores a code word that is referred to when performing encoding, and the coefficient data that determines the zero. A coefficient encoding unit that encodes coefficient data when it is determined that the coefficient data is not zero by using the code table; and when the coefficient data is determined to be zero by the zero determination unit, Zero counter means for counting the number of consecutive zeros until the zero determination unit determines that the coefficient data is not zero, and one counter for the continuous length of the zeros (hereinafter, zero run length) A table in which one table number is associated with a zero run length, a group table which manages a plurality of zero run lengths as one group and associates one table number with the table, and the continuous table counted by the zero counter means. Zero run grouping means for determining a group number corresponding to the number of zeros using the group table, the group number obtained by the zero run grouping means, and the code table, And a zero-run coding means for coding the number of zeros to be performed. 2. The variable length coding apparatus according to claim 1, wherein the group table has information in which the number of the consecutive zeros is associated with the minimum number of bits necessary for encoding the number. The zero-run encoding unit uses the group number obtained from the zero-run grouping unit, the minimum number of bits obtained from the group table, and the code table to calculate the number of consecutive zeros. The variable-length encoding device according to claim 1, wherein encoding is performed. 3. A variable variable for dividing an input signal into coefficient data representing a plurality of different frequency bands, performing quantization on the coefficient data for each of the frequency bands, and encoding the quantized coefficient data. A length encoding device, a zero determination unit that determines whether the coefficient data is zero, a coefficient table in which the coefficient data value is associated with a table number for the coefficient value, When it is determined, the coefficient table decoding means for determining the table number of the non-zero coefficient data using the coefficient table, the number of effective bits of the quantized coefficient data, and the coefficient table decoding means A non-zero coefficient encoding means for performing encoding using a table number to be encoded and the code table. . 4. The variable length coding apparatus according to claim 3, wherein the code table used for coding the non-zero coefficient data has a table size of 8 to 16. Device. 5. A variable variable for dividing an input signal into coefficient data representing a plurality of different frequency bands, performing quantization on the coefficient data for each of the frequency bands, and encoding the quantized coefficient data. A length encoding device, a zero determination means for determining whether or not the coefficient data is zero, a table number being associated with one coefficient absolute value for the absolute value of the coefficient, Absolute value is 1
A coefficient group table for managing one table number in association with one group, and a non-zero coefficient data when the zero determination unit determines that the coefficient data is non-zero. And a group encoding unit for performing encoding using the group number obtained from the coefficient group decoding unit and the code table. 6. The variable-length encoding device according to claim 5, wherein the coefficient group table manages information on the number of bits (effective number of bits) required to represent the absolute value of the coefficient for each group. A variable length coding apparatus, wherein the group coding means performs coding using information on the number of effective bits in addition to the group number obtained from the group decoding means and the code table. 7. The variable length coding apparatus according to claim 5, wherein said variable length coding apparatus has a plurality of said coefficient group tables, and selects one of said plurality of coefficient group tables. Wherein the group encoding means performs encoding using the selected coefficient group table. 8. A variable-variable device which divides an input signal into coefficient data representing a plurality of different frequency bands, quantizes the coefficient data for each of the frequency bands, and encodes the quantized coefficient data. In the long encoding method, a zero determination step of determining whether the coefficient data is zero, and storing a code word of the coefficient data when the coefficient data is determined not to be zero by the zero determination unit. A coefficient encoding step of performing encoding using the code table, and when the coefficient data is determined to be zero by the zero determination unit, the zero determination unit determines that the coefficient data is not zero. A zero count step of counting the number of consecutive zeros until the length of the zero continues (hereinafter referred to as a zero run length). A reference is made to a group table that manages a plurality of zero run lengths as one group and associates one table number with the table number, and corresponds to the number of consecutive zeros counted in the zero count step. A zero-run grouping step of determining a group number; a zero-run encoding step of encoding the number of consecutive zeros using the group number and the code table obtained from the zero-run grouping step; A variable-length encoding method comprising: 9. The variable length encoding method according to claim 8, wherein information that associates the number of the consecutive zeros with the minimum number of bits necessary to encode the consecutive zeros is added to the group table. The zero-run encoding step uses the group number obtained from the zero-run grouping step, the minimum number of bits obtained from the group table, and the code table to calculate the number of consecutive zeros. A variable-length encoding method characterized by encoding. 10. A variable variable for dividing an input signal into coefficient data representing a plurality of different frequency bands, performing quantization on the coefficient data for each of the frequency bands, and encoding the quantized coefficient data. In the long encoding method, a zero determination step of determining whether the coefficient data is zero, and if the zero determination step determines that the coefficient data is non-zero, the non-zero coefficient data is replaced with the coefficient data value. A coefficient table decoding step of determining a table number using a coefficient table in which a table number is associated with the coefficient value; an effective bit number of the quantized coefficient data; and a table obtained from the coefficient table decoding step. A variable-length encoding method, comprising: performing encoding using a number and the code table. 11. The variable length coding method according to claim 10, wherein the code table used for coding the non-zero coefficient data has a table size of 8 to 16. Method. 12. A variable variable encoder that divides an input signal into coefficient data representing a plurality of different frequency bands, quantizes the coefficient data for each of the frequency bands, and encodes the quantized coefficient data. In the long encoding method, a zero determination step for determining whether the coefficient data is zero, and, when the zero determination step determines that the coefficient data is non-zero, one coefficient absolute value for the absolute value of the coefficient And a coefficient group table for managing a plurality of coefficient absolute values as one group and associating one table number with the group number of the non-zero coefficient data. The encoding is performed using the coefficient group decoding step to be determined, the group number obtained from the coefficient group decoding step, and the code table. Performing a variable-length encoding method. 13. The variable length encoding method according to claim 12, wherein information of the number of bits (effective number of bits) required to represent a coefficient absolute value for each group is added to the coefficient group table and managed. The step of performing the encoding includes encoding using the information of the number of effective bits in addition to the group number obtained from the group decoding step and the code table. Long encoding method. 14. The variable length coding method according to claim 12, wherein a plurality of the coefficient group tables are prepared, and one of the plurality of coefficient group tables is selected. A variable length encoding method, wherein the step of performing the encoding is performed using a coefficient group table selected by the coefficient group table selection.