JPH06105296A - Variable length coding and decoding method - Google Patents

Abstract

(57) [Summary] [Objective] To propose a method for expanding the range of coefficient levels prepared by a two-dimensional Huffman code table without lowering the coding efficiency. In encoding, the required accuracy of the orthogonal transform coefficient is switched according to the required image quality, thereby adaptively expanding the level range of the coefficient of the variable-length code table such as two-dimensional Huffman. In the decoding, the variable length code table such as the two-dimensional Huffman code is adaptively expanded according to the accuracy of the transmitted orthogonal transform coefficient.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture coding and decoding apparatus.

[0002]

2. Description of the Related Art Huffman coding is a typical coding method. One of them, two-dimensional Huffman coding, is widely used for transmission of transform coefficients of two-dimensional DCT (Discrete Cosine Transform), which is a typical moving picture coding system.

Transmission of transform coefficients of a two-dimensional DCT will be described with reference to FIGS. 1 and 3.

A two-dimensional DCT image is divided into 8 × 8 blocks, and each block is subjected to two-dimensional DCT to obtain DCT transform coefficients.

Quantization of Transform Coefficients DCT transform coefficients are quantized at the quantization step given by the quantizer.

Zigzag scanning The quantized transform coefficients are stored in a one-dimensional array while performing zigzag scanning as shown in FIG.

Two-dimensional Huffman Coding A quantized coefficient stored in a one-dimensional array is examined in this order, and for a conversion coefficient having a non-zero value, its value and relative position are set to form a variable length code. And transmit.
First, find the number of transform coefficients (insignificant coefficients) with a quantized output of zero sandwiched between the significant transform coefficient to be coded and the preceding significant transform coefficient that has already been coded, and then calculate the run length. (Called run-length). The combination (run-l) of this run length and the value of the significant conversion coefficient to be encoded (called level)
ength, level) based on a two-dimensional Huffman code table (Table 1) prepared in advance. And if you have coded all the significant coefficients in the block, then
A code called EOB (end of block) is sent using the Huffman code in Table 1.

The above is the Huffman coding algorithm.

Next, the two-dimensional Huffman code table of Tables 1 to 5 will be described. Tables 1 to 5 are ISO / IEC
/ JTC1 / SC2 / WG11 (commonly called MPEG)
, A standard method of moving image coding that has been decided (commonly known as M
2 is a two-dimensional Huffman code table for coding DCT coefficients used in (referred to as PEG1).

[0010]

[Table 1]

[0011]

[Table 2]

[0012]

[Table 3]

[0013]

[Table 4]

[0014]

[Table 5]

In the tables of Tables 1 to 4, the second column is a conversion coefficient (non-conversion) having a zero quantized output sandwiched between the significant conversion coefficient to be encoded and the immediately preceding significant conversion coefficient already encoded. It represents the number of significant coefficients, that is, the run length (called run-length), and the third column represents the absolute value (called level) of the significant conversion coefficient to be encoded. The first column is the Huffman code corresponding to the above (run-length, level) combination.

The two-dimensional Huffman code table has all run lengths (0 to 63) and all levels (-255 to +).
No VLC (variable length code) code is prepared for the combination (255), and the FLC (fixed length code) code shown in Table 5 for the combinations of run length and level which are considered to be infrequently used. Is prepared.

The given combination of run length and level is shown in Table 1.
If it does not exist in the table of ~ 5, the variable length coder first uses "e
The "scape" code (Table 1) is output, then the run length is output as a 6-bit FLC code (Table 5), and then the level is output based on Table 5.

First, when the level is in the range of -127 to +127, an 8-bit FLC code is output. If the level is out of this range, the 8-bit "escape" code ("0000 0000" or "1000 0000") is output and then 8
The level is output with a bit FLC code.

The two-dimensional Huffman code table used in the Huffman code is not always prepared for all the values of the significant transform coefficients, which may cause a problem. For example, in MPEG1, the range of the pixel value of the input image is 9 bits (-255 to +255), so the range of the two-dimensional DCT conversion coefficient is 12 bits (-2048).
To +2047).

However, the two-dimensional Huffman code table used in MPEG1 referred to in Tables 1 to 5 is
Is prepared only in the range of -255 to +255. Therefore, at the quantization stage in FIG. 1, it is necessary to monitor the quantized coefficient values so that the quantized coefficient values fall within the range of −255 to +255. If the quantized coefficient value exceeds this range, the quantization step must be further increased so that it falls within the range.

Therefore, in a block in which a transform coefficient of such a large level has occurred, a quantization step larger than initially desired must be used, and the image quality must be lowered.

In view of such a current situation, it is considered that the range of the coefficient level prepared by the two-dimensional Huffman code table needs to be expanded in order to achieve higher image quality in MPEG1. No specific method has yet been presented for.

[0023]

[Problems to be Solved by the Invention] Conventionally, -255 to +
When a two-dimensional Huffman code table prepared only for the range of 255 is expanded to a larger range, simply expanding the maximum range to 12 bits (-2048 to +2047) causes waste of coding efficiency. .

That is, even if the required image quality does not require 12 bits as the precision of the quantized DCT coefficient, if only a table extended to the 12-bit range is prepared as a table, a redundant code is used. It will be output. Therefore, when expanding the two-dimensional Huffman code table, it is necessary to devise a table expansion of the number of bits according to the required image quality.

[0025]

On the encoder side, the accuracy (the number of bits) of an orthogonal transform coefficient such as DCT required according to the required image quality is sequenced or GOP (gr
oup of Pictures) or pictures or slices, or MB (macro block) or block units, and the variable length encoder adaptively extends the variable length code table such as the two-dimensional Huffman code according to the information. And output a variable length code.

On the decoder side, the transmitted DCT
Sequences the accuracy (number of bits) of orthogonal transform coefficients, such as
Or, a GOP (group of Pictures), a picture, or a slice, or an MB (macro block) or a block unit is received, and accordingly, the decoder adaptively extends a variable length code table such as a two-dimensional Huffman code. Is performed and the transmitted orthogonal transform coefficient such as DCT is decoded.

[0027]

According to the encoder of the present invention, the variable length table is adaptively expanded according to the required accuracy (bit number) of the orthogonal transform coefficient such as DCT according to the required image quality. Since the long code is output, it is possible to perform coding without waste.

Further, according to the decoder of the present invention, the table of the variable length code such as the two-dimensional Huffman code is adaptively expanded and decoded according to the accuracy of the orthogonal transform coefficient such as the DCT transmitted. Therefore, it is possible to perform decoding without waste.

[0029]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Before starting the description of the embodiments, first, sequences, GOPs, pictures, slices, MBs (macro blocks), and blocks will be described with reference to FIG.

Block Layer The block is composed of, for example, 8 lines × 8 pixels adjacent to each other in luminance or color difference. For example, DCT is executed in this unit.

MB layer MB is composed of four luminance blocks adjacent to each other on the left, right, top and bottom, and a total of six color difference blocks of Cb and Cr corresponding to the same position on the image. The order of transmission is Y0, Y1, Y2, Y3, Cb, Cr. What is used for the prediction data, whether it is not necessary to send the difference, and the like are determined in this unit.

The slice layer image is composed of one or a plurality of macroblocks arranged in the scanning order. At the head of the slice, the difference between the motion vector and the DC component in the image is reset, and the first macroblock has the data indicating the position in the image, so that it is possible to recover even if an error occurs. There is. Therefore, the length and starting position of the slice are arbitrary and can be changed depending on the error condition of the transmission path.

PICTURE LAYER PICTURE, that is, each image is composed of at least one or more slices. Then, it is classified into I-picture, P-picture, B-picture and D-picture according to the encoding method.

GOP A GOP is composed of one or more I-pictures and zero or more non-I-pictures.

Video Sequence Layer A video sequence is composed of one or more GOPs having the same image size, image rate and the like.

(1) First Embodiment In the first embodiment, the case of expanding the coefficient level range of the two-dimensional Huffman code table for DCT coefficients used in MPEG1 will be described.

The operation of the encoder will be described with reference to FIG. Step (e-1) Depending on the required image quality, a signal representing the required accuracy (number of bits) of the orthogonal transform coefficient such as DCT is sequenced, or GOP (group of Pictures), picture, or slice, Alternatively, it is transmitted in MB (macro block) or block units.

For example, when the precision is 8 bits, -127 to +127 9 bits, -255 to +255 10 bits, -511 to +511 11 bits, -1023 to. In the case of +1023 12 bits, the level range of -2047 to +2047 can be expressed.

Step (e-2) It is checked whether or not the combination of run-length and level to be encoded exists in the VLC table (Tables 1 to 4). If so, it outputs the applicable VLC code. If not, proceed to step (e-3).

Step (e-3) If the combination of run-length and level to be encoded does not exist in the VLC table, first output the "escape" code (Table 1), then refer to Table 5, The run length is output as a 6-bit FLC code (Table 5).

Step (e-4) Referring to Table 6 on the basis of the signal indicating the accuracy (the number of bits) of the orthogonal transform coefficient in step (e-1), the level is output with the bit width.

[0042]

[Table 6]

The operation of the decoding device will be described with reference to FIG. Step (d-1) sequence, GOP (group of Pictures), picture, slice, or MB (macro bl)
ock) or a signal representing the precision (the number of bits) of the DCT transform coefficient transmitted in block units.

Step (d-2) The transmitted two-dimensional Huffman code is decoded. run-
If the length is not an "escape" code (Table 1), get the decoded run-length and level values. Otherwise, go to step (d-3).

Step (d-3) If the run-length is the "escape" code (Table 1), then referring to Table 5, read the 6-bit run-length.
The value obtained is the run-length.

Step (d-4) Based on the signal indicating the accuracy (the number of bits) of the orthogonal transform coefficient in step (d-1), refer to Table 6 and read the level in the number of bits. The obtained value becomes the level.

(2) Second Embodiment Similar to the first embodiment, the DCT used in MPEG1
A case of expanding the coefficient range of the two-dimensional Huffman code table for coefficients will be described. The operation of the encoding device will be described based on FIG.

Step (e-1) A signal representing the required accuracy (the number of bits) of the orthogonal transform coefficient such as DCT is sequenced, or GOP (group of Pictures), or picture, according to the required image quality. Alternatively, it is transmitted in units of slices, MBs (macro blocks), or blocks.

Step (e-2) It is checked whether or not the combination of run-length and level to be encoded exists in the VLC table (Tables 1 to 4). If so, it outputs the applicable VLC code. If not, proceed to step (e-3).

Step (e-3) If the combination of run-length and level to be encoded does not exist in the VLC table, first output the "escape" code (Table 1), then refer to Table 5, The run length is output as a 6-bit FLC code (Table 5).

When the step (e-4) level is in the range of -127 to +127, the 8-bit FLC code is output. If not, proceed to step (e-5).

Step (e-5) When the level is a positive value, the code "0000 0000"
When the value is negative, the code "1000 0000" is output. Then, referring to Table 7, based on the signal representing the accuracy (the number of bits) of the orthogonal transform coefficient in step (e-1),
The level is output with a width of (the number of bits-1).

[0053]

[Table 7]

The operation of the decoding device will be described with reference to FIG. Step (d-1) sequence, GOP (group of Pictures), picture, slice, or MB (macro bl)
ock) or a signal representing the precision (the number of bits) of the DCT transform coefficient transmitted in block units.

Step (d-2) The transmitted two-dimensional Huffman code is decoded. run-
If the length is not an "escape" code (Table 1), get the decoded run-length and level values. If not,
Proceed to step (d-3).

Step (d-3) If the run-length is the "escape" code (Table 1), then referring to Table 5, read the 6-bit run-length.
The value obtained is the run-length.

Step (d-4) Next, 8 bits are read. The value at this time is "0000 000
If it is not 0 "or" 1000 0000 ", this value is set as the level. If not, proceed to (d-5).

If the step (d-5) value is "0000 0000" or "1000 0000", then, based on the signal representing the precision (the number of bits) of the orthogonal transform coefficient at step (d-1), Referring to Table 7, the level is read with a width of (the number of bits-1). The obtained value becomes the level.

By the way, the sequence or G
A method for changing the precision (the number of bits) of a transform coefficient such as an OP (group of pictures), a picture, a slice, an MB (macro block), or a DCT transmitted in block units will be described.

(A) When the accuracy is determined before the encoding (a) When the encoding is performed in real time The accuracy of the transform coefficient such as DCT is determined in advance in accordance with the required image quality, and it is necessary accordingly. According to the above, expansion of the variable length table for coefficients is instructed in sequence units. For example, the maximum range of the DCT coefficient level is -2047 to +20.
If it is 47 (this is the maximum range output from the MPEG DCT module) and you want Loss-Less coding as the required image quality, the level of the variable length table for the required coefficients. For the range of, a maximum of 12 bits must be prepared.

(B) When encoding in non-real time First, the DCT is previously performed on the moving image to be encoded.
Orthogonal transformation such as is performed, and the statistical distribution of the transformation coefficient level is investigated. Then, based on this statistical data and the required image quality, the level range of the variable length table for the required coefficient is determined. For example, when the maximum level of the above statistical data is 1000 and the desired image quality is Loss-Less coding, the level range of the variable length table for the required coefficient is , -1023 to +1023 (11 bits) are required.

The above statistical data is sequenced or G
By investigating in units of OP, picture, slice, or MB, the range of the level of the variable length table for coefficients is adaptively determined based on the respective statistical data. When initializing the level range of the variable length table or changing it during processing, a sequence, or GOP, or picture, or slice, or M
This is indicated by a 1-bit flag that exists in B units,
Then, information indicating the level range of the variable length table to be used is transmitted.

(B) Case where accuracy is changed during encoding This is a case where the accuracy is changed in MB units or block units while a moving image is being encoded in real time. After the image data input in MB units is subjected to orthogonal transformation such as DCT and the coefficients are quantized, the maximum absolute value of the quantized value is determined, and the maximum value is initially designated as a variable length. If the level range of the table is exceeded, the encoder indicates to change the level range of the variable length table.

For example, the level range of the initially designated variable length table is 9 bits (-255 to +255).
When the maximum absolute value of the quantized value is 300, the encoder sets the level range of the variable length table to 1
Instruct to set to 0 bit. If the maximum absolute value of the quantized value falls within the range of 9 bits (-255 to +255) when processing MBs of the addresses after that, the encoder determines the level of the variable length table. Instructs to return the range of to 9 bits again.

When changing the level range of the variable length table during processing, a 1-bit flag existing in the MB or block unit is used to indicate that, and information indicating the level range of the variable length table to be used next. To transmit.

As described above, according to the first embodiment or the second embodiment, the D required according to the required image quality on the encoder side.
According to the accuracy (number of bits) of the orthogonal transform coefficient such as CT,
It is possible to output the variable length code by adaptively expanding the variable length code table.

On the decoder side, the transmitted DC
Depending on the accuracy of the orthogonal transform coefficient such as T, the variable length code table such as the two-dimensional Huffman code is adaptively expanded, and the coefficient can be decoded.

[0068]

According to the present invention, the encoder side switches the required accuracy (the number of bits) of the orthogonal transform coefficient such as DCT according to the required image quality, thereby changing the variable length such as two-dimensional Huffman. Since the level range of the coefficient of the code table can be adaptively expanded, efficient coding is possible.

On the decoder side, the transmitted DC
Depending on the accuracy of the orthogonal transform coefficient such as T, the variable length code table such as the two-dimensional Huffman code is adaptively expanded, and the coefficient can be decoded.

[Brief description of drawings]

FIG. 1 is a diagram showing a method for encoding and decoding orthogonal transform coefficients.

FIG. 2 is a diagram showing a zigzag scan order of orthogonal transform coefficients.

FIG. 3 is a diagram showing a method of encoding and decoding a DCT coefficient.

FIG. 4 is a diagram showing an algorithm for encoding a variable length code according to the first embodiment.

FIG. 5 is a diagram showing an algorithm for decoding a variable-length code according to the first embodiment.

FIG. 6 is a diagram illustrating an algorithm for encoding a variable length code according to the second embodiment.

FIG. 7 is a diagram illustrating an algorithm for decoding a variable-length code according to the second embodiment.

Claims (6)

[Claims] 1. The required accuracy of the orthogonal transform coefficient is sequenced or GOP (grou (grou) according to the required image quality.
p of Pictures), or pictures, or slices,
Alternatively, the variable length coding method is characterized in that the variable length coding table is transmitted in MB (macro block) or block units, the variable length coding table is adaptively expanded according to the information, and the variable length coding is output. 2. The accuracy of transmitted orthogonal transform coefficients is sequence, GOP (group of Pictures), picture, slice, or MB (macro bloc).
k) or in block units, adaptively expanding the variable-length code table accordingly, and decoding the transmitted orthogonal transform coefficient. 3. A signal, which represents the required accuracy of the orthogonal transform coefficient, is sequenced or G depending on the required image quality.
OP (group of Pictures), or picture or slice, or MB (macro block) or block is transmitted, and it is checked whether or not a combination of run-length and level to be encoded exists in the VLC table and exists. In this case, the applicable VLC code is output, and if the combination of run-length and level to be encoded does not exist in the VLC table, first the "escape" code is output, and then the run length is the FLC code of the given bit. A variable-length coding method which outputs the level and outputs the level in the bit width based on a signal indicating the accuracy of the orthogonal transform coefficient. 4. A sequence or GOP (group of P)
ictures), pictures, slices, or MB (macro block) or a signal representing the accuracy of orthogonal transform coefficients transmitted in block units, decodes the transmitted two-dimensional Huffman code, and
Is the "escape" code, the decoded run-length
And the values of level and run-length are "escape" codes, then read the run-length of the given bit,
A variable length decoding method in which the obtained value is set as run-length, the level is read in the number of bits based on the signal representing the accuracy of the orthogonal transform coefficient, and the obtained value is set as the level. 5. A signal representing the required accuracy (the number of bits) of the orthogonal transform coefficient is sequenced, or GOP (group of Pictures), or picture or slice, or MB (macro) according to the required image quality. block) or the block-unit, and the run-length to be encoded and
It is checked whether or not a level combination exists in the VLC table. If so, the applicable VLC code is output, and the combination of run-length and level to be encoded is VL.
If it does not exist in the C table, the "escape" code is first output, and then the run length is output as the FLC code of the first bit number. When the level is within the predetermined range, the second code is output.
The FLC code of the bit number of is output when the level is not within the range, the first code is output when the level is a positive value, and the second code is output when the level is a negative value. A variable length coding method, wherein the level is output in a width of (the number of bits-1) based on a signal indicating the accuracy of the orthogonal transform coefficient. 6. A sequence or GOP (group of P)
ictures), pictures, slices, or MB (macro block) or a signal representing the accuracy of orthogonal transform coefficients transmitted in block units, decodes the transmitted two-dimensional Huffman code, and
Is the "escape" code, the decoded run-length
And the value of level, and if run-length is "escape" code, then read run-length of the first number of bits, set the obtained value as run-length, and then the second number of bits If the value at this time is not the first code or the second code, this value is set as the level, and if the value is the first code or the second code, then the step A variable-length decoding method in which a level is read in a width of (the number of bits-1) based on a signal indicating, and the obtained value is used as a level. [0000]