KR0165504B1 - Signal compression coding method - Google Patents

Abstract

The present invention relates to a compression coding method of an HD-Baseband signal, and calculates the total number of zero zero coefficients of one segment, divides the total number of zero zero coefficients by a predetermined number, and determines the quantization number primarily. Classify the discrete cosine transform blocks into the first to fourth classes, classify the discrete cosine blocks into the first to eighth regions, and perform variable length coding on all quantized values constituting one segment by zigzag scanning. A predetermined bit is assigned to each macroblock by bits generated by encoding in the first variable length encoding step, and a difference between the allocated bit and the generated bit is obtained to determine the quantization number of each macroblock. According to the first order quantization number adjustment table of N, performing a second variable length encoding, and encoding a predetermined bit for each macroblock by the generated bits. Allocating bits, and obtaining the difference between the generated bits and the bit allocation comprises a step of adjusting in accordance with the quantization number subtraction of each macroblock in the predetermined second number of quantization adjustment table. Accordingly, by repeatedly adjusting the quantization width, a method is provided in which the lengths of the code bits are close to each other in signal compression so that they can be most efficiently utilized with a given fixed length bit. This provides an effect of improving image quality.

Description

Signal compression coding method

1 is a diagram showing a frame structure and a macro block structure of an HD-Baseband signal.

2 is a diagram showing the structure of macroblock shuffling and segments.

3 is a schematic diagram illustrating a signal compression method according to the present invention.

4 is a quantization table according to an embodiment of the present invention.

5 is a diagram illustrating class division according to an embodiment of the present invention.

6A and 6B are views for explaining the area divisions in the DCT block.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compression encoding method of an HD-Baseband (High Definition-Baseband) signal, and more particularly, to a method for generating a quantization width of a bit amount generated in a quantization step.

In order to record a digital video signal on a high-definition video cassette recorder (HD-VCR), a method of fixed compression in a predetermined area unit is essential. At present, the international standard for recording method of HD-Baseband signal is decided, and most specifications are fixed in consultation and some parts can be produced according to the manufacturer's specification. Therefore, the manufacturer's specifications dominate the overall system performance. As a result, only the decoding process is determined and it is only necessary to maintain compatibility that the signals encoded by any method can be decoded.

In recording a digital video signal on a videotape, a variable number coding method with a constant output occurs because a certain number of tracks is occupied every frame, which causes a problem when recording a compressed video signal. That is, since the output is not constant, the amount of bits generated and occupied according to a given image is variable, and the compressed and output image signal may be smaller or larger than the fixed target length. In this case, the code length output method that is closest to the target bit length is required because it can greatly affect the image quality to be reproduced.

In order to realize such fixed length encoding, the encoding length is fixed in one segment unit composed of a predetermined number of macroblocks. In the method of enabling independent decoding by one segment unit, the most important part which greatly influences the performance is the quantization process. As a result, the quantization process determines the most suitable quantization width, and the amount of bits generated according to the degree of complexity in units of macroblocks is different.

Accordingly, an object of the present invention was devised to satisfy the above-mentioned proposal, and the five macroblocks form one segment through shuffling, and the AC coefficient generated in the process of quantizing the DCT transformed AC coefficients. Determining the initial quantization width with respect to the input signal, and repeatedly adjusting the quantization width using the difference between the bit and the target bit generated close to the target code length for the code length generated through the variable length encoding process To provide a method for determining the quantization width.

In the signal compression method performed by determining an efficient quantization width for achieving the above object, in a method of encoding a plurality of macro blocks composed of a predetermined DCT block by one segment unit,

A first step of calculating the number of AC coefficients of the luminance signal of the DCT block constituting the one segment being non-zero and dividing the calculated number by a predetermined value to determine a quantization number; Classifying all DCT blocks forming one segment into a plurality of classes according to the sum of absolute values of the AC coefficients, and classifying each DCT block into a plurality of regions according to frequency components of the AC coefficients; A third step of setting a quantization width according to the quantization number determined in the first step and the classes and regions classified in the second step; A fourth step of variable length encoding the values obtained by dividing the respective AC coefficients for all the discrete cosine transform blocks constituting one segment by the quantization width set in the third step; Assigning a predetermined bit to each macroblock by the bit coded and generated in the fourth step, and adjusting the quantization number determined in the second step by the difference between the allocated bit and the bit actually generated. Step 5; The quantization width is set according to the quantization number adjusted in the fifth step and the class and region classified in the second step, and each AC coefficient is applied to all the discrete cosine transform blocks constituting one segment with the set quantization width. A sixth step of variable length encoding the value obtained by dividing the sound; And allocating predetermined bits for each macroblock by the bits generated by encoding in the sixth step, and re-adjusting the quantization number adjusted in the sixth step by the difference between the allocated bits and the bits that actually occur. And a seventh step of adjusting.

Hereinafter, with reference to the accompanying drawings will be described the present invention in more detail.

1 is a diagram showing a frame structure and a macro block structure of an HD-Baseband signal. The input image is 1008 x 1024. The sampling format is 12: 4: 0, and one macro block is composed of eight DCT blocks each composed of two chrominance components with respect to six luminance signals, and becomes a basic unit of block shuffling quantization stem determination.

2 is a diagram showing the structure of macroblock shuffling and segments.

One segment is constructed by taking five macro blocks from various locations on the screen. The segment consisting of five macroblocks can be efficiently fixed in the amount of information of the block as the information amount of the building block is constant. Therefore, the macroblock unit is mixed with respect to the entire frame. Therefore, the fixation of the code amount is performed in units of one segment. Shuffling consists of five macro blocks. Typically, C is shuffled to MB1, B to MB2, D to MB3, A to MB4, and E to MB5.

In the HD-Baseband standard, DCT conversion is divided into two modes according to the characteristics of the signal. The two DCT modes refer to frames and fields, respectively, to increase DCT efficiency.

The selection of the DCT mode is determined in units of 8x8 blocks. Therefore, in the case of an image having a lot of motion, the selection of the field mode is increased. In the DCT mode, the 8x8 DCT mode is selected when the value between the two fields is less than the predetermined reference value, and the 2x4x8 DCT mode is selected when the DCT mode is large.

3 is a schematic diagram illustrating a signal compression method according to the present invention.

Each macro block having a discrete cosine transform (DCT) block of six luminance signals (Y) and a discrete cosine transform (DCT) block of two color difference signals (Cr, Cb) is one as five micro blocks through shuffling. Constitute a segment of. The segment unit becomes a basic unit of compression encoding. In the step 300 of calculating the total number of nonzero coefficients in each segment, the number of nonzero coefficients (nonzero coefficients) is calculated for all AC coefficients of the DCT block of six luminance signals constituting one segment. . Here, a coefficient other than 0 (zero) means a number equal to or greater than absolute value 1. That is, the zero coefficient means a number smaller than the absolute value 1. In the first quantization number determination step 302, the number of nonzero coefficients calculated in the nonzero coefficient count calculation step 300 is divided by a predetermined jet number to determine the quantized value as the quantization number. Here, the meaning of the predetermined number of jets is as follows.

In HD-Baseband, a segment unit, which is a basic unit of compression encoding of a signal, is composed of five macro blocks. In addition, one macro block includes six luminance signal DCT blocks, and each DCT block is composed of 64 numbers as 8 × 8. The 64 coefficients consist of one DC coefficient and 63 AC coefficients. The 64 coefficients consist of one DC coefficient and 63 AC coefficients. Then, the total number of AC coefficients in one segment unit is 5 × 6 × 63 = 1890. That is, the maximum value of the number of nonzero coefficients of the AC coefficient in the compression encoding of the signal is 1890. Of course, the minimum value of the number of nonzero coefficients of the AC coefficient is 0 (zero). In other words, the number of AC coefficients in the segment unit is limited to the range of 0 (zero) to 1890. On the other hand, the number Q of quantization numbers QNO that can be taken respectively is specified. Then, the setting of the initial quantization number (QNO) in the first quantization number determination step 302 according to the present invention should be able to embrace the number of AC coefficients in all cases that can occur in the one segment unit. That is, a value obtained by dividing the total number of nonzero coefficients of the AC coefficients available in the segment unit by the number Q of the quantization number QNO is set to the predetermined jet number. In the present invention, for convenience of calculation, the number of maximum nonzero coefficients of the AC coefficient in segment units is set to 5x6x64 = 1920. Here, multiplying by 64 includes the number of DC coefficients. Although AC and DC coefficients are performed separately in the compression encoding of the actual signal, it is clear that the calculation is performed as 64 for convenience of calculation. In the present invention, the number Q of quantization numbers QNO is 16. That is, the quantization number QNO is set to Q1, Q2 ...., Q16. Then, dividing the maximum number of nonzero coefficients of the AC coefficient in units of segments by 16 may occur in one segment unit in setting an initial quantization number (QNO) in the first quantization number determination step 302 of the present invention. The number of AC coefficients in all cases can be embraced. Then, dividing 1920 by 16 gives 120. That is, as an embodiment of the present invention, in setting the initial quantization number QNO in the first quantization number determination step 302, the number of AC coefficients in all cases that may occur in one segment unit may be included. Then, dividing 1920 by 16 gives 120. That is, as an embodiment of the present invention, in the first quantization number determination step 302, the number of nonzero coefficients calculated in the number of nonzero coefficient counting step 300 is divided by 120 to determine a quantized value as a quantization number. do.

In fact, almost no AC coefficient becomes zero in one segment unit, and even if all AC coefficients are zero, it is not necessarily limited to 120. However, if possible, the initial quantization number (QNO) is set by setting a predetermined number of jets to allow the case where all the AC coefficients can be obtained.

In the class and region classification step 304, all DCT blocks constituting one segment according to the sum of the absolute values of the respective AC coefficients in each DCT block converted by the DCT are first, second, third, and Classify as a fourth class and classify in each DCT block. In the quantization step, the quantization width varies depending on the quantization number (QNO), the class, and the area. Quantization width can be found at a glance using a quantization table. An embodiment of the quantization table will be described with reference to the drawings.

4 is a quantization table according to an embodiment of the present invention. 4 is described after explaining FIG. 5, FIG. 6A, and FIG. 6B.

5 is a diagram illustrating class division according to an embodiment of the present invention. As described above, classes are classified into first to fourth classes according to the sum of absolute values of respective AC coefficients in the DCT block. If the sum of the absolute values of the respective AC coefficients of the predetermined DCT block in the luminance signal is 6, the DCT block corresponds to class 0.

6A and 6B are diagrams illustrating the division of areas in a DCT block. FIG. 6a shows an AC coefficient divided into 0 to 7 areas according to frequency components in an 8 × 8 DCT block, and FIG. 6b also illustrates an AC coefficient into an area 0 through 7 according to frequency components in a 2 × 4 × 8 DCT block. Divided.

As shown in FIG. 4, the quantization width for dividing the AC coefficients varies according to the quantization number class and the area. In FIG. 4, classes are divided into 0, 1, 2, and 3, and 0 to 15 numbers represent quantization numbers. In addition, the areas are divided from 0 to 7, and numerical values of 1,2,4,8 and 16 mean quantization widths. That is, it divides the value of AC coefficient. As an example, the predetermined AC coefficient value in a given DCT block is 8, the quantization number QNO is 7 (400), the class is 0 (402), and the area is 3 (404). ), The quantization width is 2 (406).

Then, the value 8 of the AC coefficient is divided by the quantization width 2. Eventually, the value of 4 is the AC coefficient value is quantized.

In FIG. 3, in the first variable length coding step 306, the quantization number QNO determined in the first quantization number determination step 302 and classified in the class and area classification step is classified. And for each of the discrete cosine transform (DCT) blocks constituting all the discrete cosine transform (DCT) segments constituting a segment with the quantization width set in a predetermined quantization table according to the class and the area. Variable length coding is performed by a zigzag scan by dividing the AC coefficients.

Variable length coding selects an appropriate code from the variable length coding table according to the RUN and AC coefficients of zigzag-scanned '0', and the macroblock buffer cannot fill the allocated bits for each block. The segment buffer takes as much code as is allocated for each macro block, and if it exceeds the buffer, it puts the bits allocated to the macro block where it cannot fill.

In the first quantization number adjustment step 308, a predetermined bit is assigned to each macroblock by bits generated by encoding in the first variable length encoding step 306, and between the allocated bits and the generated bits. The difference is calculated and adjusted according to a predetermined first quantization number adjustment table.

The quantization number adjustment table changes the quantization number QNO according to the degree of the value of the difference bit.

a = b × c ÷ d

f = c-a

a is a bit allocated for a desired bit amount in macroblock units, and b is a target code length for an AC coefficient. In HD-Baseband, 2560 bits is the target code length of the AC coefficient. c is a bit generated through variable length coding of each macroblock unit. d is a bit generated through variable length coding of one segment unit. F is the difference bit between the bit that actually occurs and the assigned bit. Firstly, bits generated through variable length coding do not properly utilize bits that can be utilized to the maximum in one segment unit.

Table 1 presents the first adjustment table of the quantization width for this adjustment.

In Table 1, E denotes a difference bit between the generated bit and the allocated bit of each macroblock unit as described above, and A, B, C, D, and E are macroblocks that form a segment as shown in FIG. . That is, in one example of Table 1, if the F value is 200 and the D macroblock corresponds to the D value, the increment of the quantization number QNO is +1.

That is, the first adjusted quantization number is first adjusted by adding the increment of the quantization number to the quantization number determined in the first quantization number determination step 302.

In the second variable length encoding step 310, the first variable length encoding is performed according to the quantization number adjusted in the first quantization number adjusting step 308 and the classes and regions classified in the class and region classification step 304. Follow the same process as the steps.

In the second quantization number adjustment step 312, predetermined bits of each macroblock are allocated by bits generated by encoding in the second variable length encoding step 310, and the predetermined bits are allocated between the allocated bits and the generated bits. The difference is obtained and adjusted according to a predetermined second quantization number adjustment table. It is the same principle as the first quantization number adjustment step 308.

For this adjustment, Table 2 presents the secondary adjustment table of quantization width.

For example, if F is -20 and the D macroblock is, the quadratic quantization number (QNO) increment is zero. In this case, it has the same number as the quantization number QNO adjusted in the first quantization number adjustment step and has a quantization width by the quantization table. The quantization width is determined using the quantization number, the class, and the region, which are second-adjusted according to Table 2, and is subjected to a third variable length encoding step (not shown). By repeatedly adjusting the quantization number, the quantization width can be set finely.

The signal compression method configured as described above provides a method of making the most efficient use with a given fixed length bit by repeatedly adjusting the quantization width so that the length of the code bit approaches the target code length in signal compression. This provides the effect of improving the image quality.

Claims (5)

In a method of encoding a plurality of macro blocks consisting of a predetermined DCT block in units of segments, calculating the number of non-zero AC coefficients of luminance signals of the DCT blocks forming one segment, and Determining a quantization number by dividing the calculated number by a predetermined value; Classifying all DCT blocks forming one segment into a plurality of classes according to the sum of absolute values of the AC coefficients, and classifying each DCT block into a plurality of regions according to frequency components of the AC coefficients; A third step of setting a quantization width according to the quantization number determined in the first step and the classes and regions classified in the second step; A fourth step of variable length encoding the values obtained by dividing the respective AC coefficients for all the discrete cosine transform blocks constituting one segment by the quantization width set in the third step; Assigning a predetermined bit to each macroblock by the bit coded and generated in the fourth step, and adjusting the quantization number determined in the second step by the difference between the allocated bit and the bit actually generated. Step 5; The quantization width is set according to the quantization number adjusted in the fifth step and the class and region classified in the second step, and each AC coefficient is applied to all the discrete cosine transform blocks constituting one segment with the set quantization width. A sixth step of variable length encoding the value obtained by dividing the sound; And allocating predetermined bits for each macroblock by the bits generated by encoding in the sixth step, and re-adjusting the quantization number adjusted in the sixth step by the difference between the allocated bits and the bits that actually occur. And a seventh step of adjusting. The signal compression method according to claim 1, wherein the first step comprises a zero coefficient when the absolute value of the AC coefficient in the discrete cosine transform (DCT) block is less than one, and a nonzero coefficient otherwise. Coding method. The signal compression encoding method of claim 1, wherein the first step determines a quantization number primarily by dividing the calculated total zero coefficient by 120. The method of claim 1, wherein the fourth and seventh steps are based on the bits generated by encoding the fifth and sixth steps. And multiplying the bits of the target code length of the AC coefficient in the segment unit and dividing the number of bits currently generated in one segment unit to allocate the bits per macro block. The signal compression encoding method according to claim 1, further comprising an n-th variable-length encoding step and an n-th quantization number adjustment step of repeatedly performing the steps having the same principle as the fourth to seventh steps.