KR100251544B1 - Apparatus and method of adaptive quantization in consideration of image complexity - Google Patents

Abstract

PURPOSE: An adaptive quantization apparatus and a method thereof considering image activity are provided to reproduce the optimum picture quality by remarkably reducing quantization distortion by a block having low activity. CONSTITUTION: A class decision element decides the class of discrete cosine transformed blocks according to the activity. A temporary storing element temporarily stores the discrete cosine transformed blocks by the segment. A data amount estimating element decides the level of macroblocks according to the number of the class of the discrete cosine transformed blocks, decides quantization according to the level, and quantizes and codes accordingly for controlling and outputting quantization number to approximate the generated data amount to arbitrary reference value. The first quantization element quantizes the discrete cosine transformed blocks by the segment stored in the temporary storing element by using the quantization number outputted from the data amount estimating element.

Description

Adaptive Quantization Apparatus and Method Considering Image Complexity

The present invention relates to a method of compressing image data, and more particularly, to adaptive quantization considering image complexity that can reduce quantization distortion by determining a quantization number according to an activity level of macroblocks between macroblocks. An apparatus and method are provided.

In general, a quantization process is essential in compressing and compressing image data. Since the conventional quantization method executes quantization buffer control for each segment or macroblock, the complexity between the macroblocks in the segment is not considered at all.

Hereinafter, a conventional video signal quantization process will be described with reference to a video signal quantization apparatus disclosed in Japanese Patent Laid-Open No. 6-78286.

1 is a block diagram of an apparatus for compressing and encoding a digital video signal. The video encoding device comprises a blocker 2 for dividing an input image into macroblocks, and a shuffling unit for forming a segment by tying macroblocks ( 4) According to the DCT mode of the motion detector 6 and the motion detector 6 which detects the motion between each field in the macroblock and determines a discrete cosine transform (hereinafter referred to as DCT) mode. A DCT unit 8 for performing DCT, an adaptive quantizer 10 for compressing a video signal in units of DCT blocks, and a variable length coding unit (hereinafter referred to as VLC) unit 22.

Normally, the input video signal is a 4: 2: 2 signal in CCIR-601 (International Radio Consultative Committee; Recommendation 601) format, and the luminance signal Y is sampled at 13.5 Mz, and the color difference signal Cb) are each sampled at 6.75 Mz.

For NTSC (National Television System Committee) and 525-line / 60Hz (scanner 525-line / field frequency 60Hz) signals, the 4: 1: 1 signal is obtained by 2: 1 subsampling the chrominance signal horizontally. For 626 line / 50Hz video signals such as Phase Alternating by Line (PAL) and Sequentiel Couleur a Memoire (SECAM), the color difference signal is subsampled in the vertical direction to 2: 1 by 4: 1: Produces a signal of zero.

The blocker 2 divides the above-described input image signal into 8 × 8 pixel blocks, and displays the same space as four 8 × 8 luminance signal Y blocks as shown in FIG. A macroblock is formed by grouping 8 × 8 color difference signals Cr and Cb. Here, FIG. 2A shows a macroblock constructed from a video signal of 626 lines / 50Hz, and FIG. 2B shows a macroblock constructed from a video signal of 525 lines / 60Hz.

In the shuffling section 4, five macroblocks located in different spaces in one frame are bundled to form a segment. The motion detector 6 determines the DCT mode by detecting the motion between each field since the 8x8 block in the macroblock consists of two field signals. The DCT unit 8 performs 8x8 DCT when the movement is small, or 2x4x8 DCT when the movement is large, according to the DCT mode.

In the DCT unit 8, an 8 × 8 DCT block that has undergone DCT by the two modes is compressed through the adaptive quantization unit 10 and the VLC unit 22, and the class information, mode information, and macro block in units of DCT blocks are compressed. Alternatively, the quantized number information in units of segments and the multiplexer 24 are multiplexed and recorded on the tape.

Here, the adaptive quantization unit 10 is composed of a buffer 12, a rating determiner 14, a data amount predictor 16, and a quantizer 18.

The class determiner 14 classifies the DCT block input from the DCT unit 8 into several classes according to the importance of the block. For example, the rating determiner 14 obtains the maximum value of the AC component of each DCT block and compares it with a threshold to classify it into four grades as shown in Table 1 below. Table 1 shows an example of the grading as presented in the SD-DVC draft. On the other hand, the rating determiner disclosed in Japanese Patent Laid-Open No. 6-78286 counts AC coefficients having a value greater than or equal to a threshold value instead of the AC maximum value of each block and performs classification using the same.

In addition, in order to prevent red color bleeding when restoring and displaying the compressed image signal again, the grade determiner 14 displays the red block detected by the red block detector 20 as shown in Table 2, and displays the red color difference signal Cr. It is used for grading. At this time, the method of detecting the red block from the pixel block in the red block detector 20 is as shown in Figure 3, which is presented in the HD-DVC draft. As a result, the Cr block estimated as the red block is quantized in more detail, thereby reducing the appearance of thin color on the reproduction screen.

The data amount predictor 16 selects a quantizer such that the amount of data compressed in units of segments is constant so that the video signal is recorded on the tape at a constant rate.

In detail, the data amount predictor 16 includes the class information of the DCT block from the class determiner 14 and the region partition information as shown in FIG. 4, and the quantization number (ie, quantization) as shown in FIG. The number of data from the number 15 to the quantization number 0) is quantized using 16 kinds of quantizers or variable length coding (VLC) to predict the amount of data when each quantizer is used. In this case, the quantization number is determined in units of macroblocks or segments. The data predictor disclosed in Japanese Patent Laid-Open No. 6-78286 obtains a data amount using 16 quantizers from Q1 to Q16, and then selects a quantizer closest to the predetermined amount among them.

The quantizer 18 quantizes DCT blocks in segments stored in the buffer 12 using a quantizer selected from the data amount predictor 16.

However, since the conventional adaptive quantization apparatus performs quantization buffer control in units of segments or macroblocks without considering the image complexity between macroblocks in a segment, image quality may be degraded due to quantization distortion during reproduction.

Accordingly, the present invention has been made in view of the above circumstances, and an object of the present invention is to provide an apparatus and method for adaptive quantization in consideration of image complexity for measuring a complexity level between macroblocks and determining a quantization number according to the level.

Another object of the present invention is to provide a quantization apparatus and method considering image complexity that can reduce quantization distortion by performing quantization buffer control considering the complexity of a macroblock.

It is still another object of the present invention to provide an adaptive quantization apparatus and method considering image complexity that can reduce quantization distortion by finely quantizing a low complexity macroblock.

1 is a block diagram showing the configuration of a conventional video encoding apparatus.

2 is a diagram showing a configuration of a macro block.

3 illustrates a method of detecting a red block.

4 is a diagram showing region division information used in the data amount predictor of FIG.

5 is a diagram showing the size of a quantization step used in the data amount predictor of FIG.

6 shows the structure of a macro block.

FIG. 7 is a diagram illustrating a bit string structure of five macro blocks constituting one segment. FIG.

8 is a block diagram showing a configuration of an adaptive quantization device according to the present invention.

9 is a block diagram showing in detail the data amount predictor of FIG.

* Explanation of symbols for main parts of the drawings

2: block part 4: shuffling part

6: food detector 8: DCT (Discrete Cosine Transform)

10: adaptive quantization unit 12, 72: buffer

14: class determiner 16, 50: data amount predictor

18, 68: quantizer 20: red block detector

22, 70: VLC (Variable Length Code) part

24: MUX (Multiplexer) 60: Level Decision Circuit

62: quantization number table 64: initial quantization number determiner

66: quantization number determiner 74: quantization number determination circuit

76: control circuit

In order to achieve the above object, the adaptive quantization apparatus considering the image complexity according to the present invention comprises: grading means for determining the class of discrete cosine transformed blocks according to the complexity; Temporary storage means for temporarily storing the discrete cosine transformed blocks in units of segments; Determine the level of the macroblock according to the number of grades of the discrete cosine transform block, determine the quantization number according to this level, and quantize and encode the quantized number to adjust the quantized number so that the generated data is close to an arbitrary reference value. Means; And a first quantization means for quantizing discrete cosine transform blocks for each segment unit stored in the temporary storage means by using the quantization number output from the data amount predicting means.

In addition, the adaptive quantization method considering the image complexity according to the present invention includes a first step of determining the number of classes of discrete cosine transformed blocks and determining the level of the macroblock according to the number of the number of classes; A second step of determining, quantizing and encoding the quantized number according to the level of the macroblock, and adjusting and determining the quantized number so that the amount of generated data is closest to an arbitrary reference value; And a third step of quantizing using the determined quantization number.

Other objects and features of the present invention other than the above object will become apparent from the description of the embodiments with reference to the accompanying drawings.

First, prior to the detailed description of the embodiment of the present invention, a bit string recording method of the SD-DVC will be described with reference to FIG. 6 and FIG.

6 shows a structure of a macro block, and FIG. 7 shows a bit string structure of five macro blocks constituting one segment.

In general, DCT blocks have different class numbers according to complexity, and each macroblock has a respective quantization number. One macroblock is composed of four Y blocks and two C blocks as shown in FIG. Here, Y blocks are 14 bytes, C blocks are 10 bytes, and variable length coding is performed, and one macro block is stored in variable length coding in units of 77 bytes. However, since each block is subjected to variable length encoding, it is impossible to encode exactly 14 bytes and 10 bytes. In other words, since each block is variable length coded, the space allocated to each block may be left or insufficient. Accordingly, the remaining bit strings written in the magnetic space are written in the remaining space of other DCT blocks grouped by the same macroblock.

7 shows a bit string recording structure of one segment composed of five macroblocks. Here, each macroblock writes a variable length coded bit string in a magnetic bit area fixed at 77 bytes. In this case, like the DCT block, each macroblock may have a remaining space or a shortage of space allocated for recording a bit string.

Therefore, the above bit strings written in each macro block will be written in the empty area of the other block.

The area allocated to the segment consisting of five macroblocks is fixed at 77 x 5 = 385 bytes, and the coded bit strings must all be recorded in this area. If the remaining bit string occurs, it is discarded. Accordingly, the amount of generated data must be controlled by predicting the amount of data in the adaptive quantizer.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to FIGS. 8 and 9.

8 shows an adaptive quantization device according to the present invention, which includes a buffer 12, a rating determiner 14, a data amount predictor 50, and a first quantizer 18. As shown in FIG. Here, the operation of the components that overlap with the conventional configuration of FIG. 1 will be briefly described.

The class determiner 14 classifies the input DCT block into four classes according to the importance of the block. The data amount predictor 50 selects a quantizer such that the amount of data compressed in units of segments is constant so that the video signal is recorded on the tape at a constant rate. Referring to FIG. 9 in detail, it is as follows.

Referring to FIG. 9, the data amount predictor 50 reads the level determining circuit 60 for determining the complexity level of the macroblock, and the quantization number table 62 storing the quantization numbers of the previous frames. An initial quantization number determiner 64 for determining the quantization number, a quantization number determiner 66 for determining the quantization number for each macroblock, a second quantizer 68 for performing quantization, and a variable length for variable length encoding of the quantized data The encoding (VLC) unit 70, a buffer 72 for storing the encoded data, and a quantization for selectively outputting the quantization number to the first quantizer 16 or the control circuit 76 by comparing the data amount with a predetermined amount. A number determination circuit 74 and a control circuit 76 for controlling the quantization number determiner 66 to adjust the quantization number are provided.

Each macroblock typically has a different image complexity. Here, although a quantization error is not noticeable even if a block having a high complexity is encoded with a large quantization number, a quantization error can be reduced only by encoding a small block with a small quantization number. Therefore, the data quantity predictor 50 first determines the complexity level in units of macroblocks using the number of classes of each DCT block, and then determines the quantization number of each macroblock in units of segments using this complexity level to optimize Enable to reproduce picture quality.

In detail, the level determining circuit 60 shown in FIG. 9 determines the complexity level of the macroblock using the class of each DCT block determined from the class determiner 14 of FIG. In this case, the level determining circuit 60 determines the level using only the number of grades assigned to four Y blocks among four Y blocks and two C blocks constituting one macroblock.

Referring to Table 3, the number of ranks of each DCT block is allocated to four of 0, 1, 2, and 3 in order of low complexity. The level determining circuit 60 determines the level of this macroblock as 0 when there are four DCT blocks having a rank number of 0 as in 1), and 1 block having a rank number of 0 as shown in 2). If there are two blocks and one block with 3, the level is determined as 12. In this way, all combinations of the number of grades that can be allocated to four Y blocks are shown in Table 3, and in each case, a total of 32 macroblock levels are determined by allocating macroblock levels. At this time, like the number of grades, the lower the complexity level of the macroblock, the lower the number is assigned.

The quantization number table 62 stores the quantization number in macroblock units of the previous frame. This is to determine the quantization number of the current macroblock more quickly and accurately by using the quantization number table of the previous frame because the image data is highly correlated in time. For example, since one frame is composed of 1350 macroblocks and the quantization number can be recorded by 4 bits (quantization numbers: 0 to 15), the quantization table 62 needs a total of 1350 x 4 bits of memory.

Since there are five macroblocks in one segment, five macroblock levels are determined first. The lowest level macroblock among the five macroblocks in one segment is called a priority macroblock, and when there are two or more lowest level macroblocks, priority is arbitrarily determined. Thus, the quantization size of the low level macroblock is determined first and the quantization size of the remaining macroblocks is determined. This is to finely encode a low complexity block.

The initial quantization number determiner 64 determines the initial quantization number of this priority block.

This reads and uses the quantization number of the previous frame macroblock at the same position in the quantization number table 62. If the frame being encoded is the first frame, 8, which is an initial value of the previous frame quantization table, is used. As such, since the quantized number value of the previous frame is used, an appropriate quantized number value can be quickly found.

When the quantization number of the priority macroblock is determined in the initial quantization number determiner 64, the quantization number determiner 66 determines the quantization number of the remaining four macroblocks in proportion to the level of the macroblock determined by the level determining circuit 60. . In other words, since the level of the macroblock is 32 levels, the quantization number of each macroblock is determined by decreasing the quantization number by one each time the level goes down by two levels compared to the priority macroblock.

For example, assume that the levels of the five macroblocks in the level determination circuit 60 are set in the order of 9, 2, 0, 21, 29. The priority macroblock is the third macroblock with the lowest level set to zero. The initial quantization number determiner 64 reads the quantization number table 62 and determines that the initial quantization number of this block is 1 if the quantization number of the previous frame macroblock, which is the same position as the current priority macroblock, is 1. Next, the quantization number determiner 66 determines that the macroblock having the level 9 is 9, which is 4 levels below the quantization frame, since the level difference from the priority macroblock is 9. Accordingly, the determined initial quantization numbers of the five macroblocks are 5, 2, 1, 11, and 15, respectively. In this case, when the level of the macroblock is 30 or 31, if the quantization number is reduced by the level difference, the quantization number becomes 16. However, since the quantization number exceeds the range (0 to 15) of the quantization number, the quantization number is determined to be 15.

In this way, after determining the quantization number of five macroblocks, the second quantizer 68 is quantized using the quantization number, and the VLC unit 70 performs a scan and variable length encoding to perform a quantization on the buffer 72. Save it.

The quantization number determination circuit 74 determines whether or not the bit amount of the five macroblocks encoded as described above is closest to the space allocated to one segment, that is, 385 bytes.

When the encoded bit amount is closest to 385 bytes, the quantization number determination circuit 74 outputs the determined quantization number to the first quantizer 16 for encoding, and converts the value to the previous frame quantization table 62. To record. On the other hand, when the coded bit amount is not close to 385 bytes, the quantization number determination circuit 74 outputs the coded data to the control circuit 76.

The control circuit 76 controls the quantization number determiner 66 to adjust the quantization number of each macro block. In other words, the control circuit 76 compares the data amount with a predetermined data amount (385 bytes) and adjusts the quantization number of each macroblock by the proportional difference. Specifically, if the encoded bit amount is largely different from 385 Byte, the control circuit 76 adjusts the quantization number of the priority macroblock. Accordingly, since the quantization numbers of the remaining four macroblocks are all changed, the amount of generated bits is greatly changed. At this time, the size of the quantization number to be adjusted is determined so as to be proportional to the difference from 385 bytes in the control circuit 76. On the other hand, if the coded bit amount is not significantly different from 385 Byte, and thus it is not necessary to adjust all the quantization numbers of the five macroblocks, the quantization number of each macroblock is adjusted by the following two methods. If the bit generation amount is smaller than 385 bytes, the quantization number is reduced from the priority macroblock having the lowest complexity level. If the bit generation amount is small even though the number of quantization numbers of the priority macroblock is reduced by one step, it is determined whether the next level is closest to 385 bytes while decreasing the number of quantization numbers in descending order. On the other hand, when the bit generation amount is larger than 385 bytes, the bit generation amount is controlled by increasing the quantization number from the macroblock having a high complexity level. This is to ensure the highest quality of the priority macroblocks. This is because a block having a low complexity can obtain a small picture quality with less distortion due to quantization error only if the quantization is finer.

Returning to FIG. 8, the first quantizer 18 quantizes the current segment stored in the buffer 12 in units of DCT blocks using the quantizer selected by the above procedure from the data quantity predictor 50.

As such, the adaptive quantizer of the present invention can reduce distortion for quantization error by finely quantizing low complexity blocks.

As described above, the adaptive quantization apparatus and method considering the image complexity according to the present invention measures the complexity level between macroblocks in a segment and determines the quantization number of each block based on the low-level block to quantize the block. Blocks are finely quantized. As a result, the quantization distortion due to the low complexity block is remarkably reduced, so that it can be reproduced at an optimum picture quality.

Those skilled in the art will appreciate that various changes and modifications can be made without departing from the technical spirit of the present invention. Therefore, the technical scope of the present invention should not be limited to the contents described in the detailed description of the specification but should be defined by the claims.

Claims (12)

An adaptive quantization apparatus of a video encoding apparatus, comprising: grade determining means for determining a grade of discrete cosine transformed blocks according to complexity; Temporary storage means for temporarily storing the discrete cosine transformed blocks in units of segments; Determining the level of the macroblock according to the number of grades of the discrete cosine transform block, and determining the quantization number according to the level, and quantizing and encoding the data to adjust the quantization number so that the generated data is close to an arbitrary reference value. Prediction means; And first quantization means for quantizing discrete cosine transform blocks for each segment unit stored in the temporary storage means by using the quantization number output from the data amount predicting means. 2. The apparatus of claim 1, wherein the data amount estimating means comprises: macroblock level determining means for determining a complexity level of the macroblock using the number of grades of the discrete cosine transform block; A quantization number table for storing quantum times of previous frames for each macroblock; Initial quantization number determination means for reading the quantization number table and determining the macroblock quantization number at the same position as the initial quantization number of the priority macroblock; Quantization number determination means for determining the quantization number of the remaining blocks by adding or subtracting in proportion to the difference between levels based on the quantization number of the priority macroblock; Second quantization means for performing quantization using the determined quantization number; Encoding means for scanning the quantized data and performing variable length encoding; A buffer for temporarily storing the encoded data; Switching means for outputting a quantized number or outputting encoded data by comparing the stored data amount with a reference value; And control means for controlling the quantization number determining means to adjust the quantization number according to the difference between the amount of data output from the switching means and the reference value. The adaptive quantization apparatus of claim 2, wherein the macroblock level determining unit determines the complexity level by using the number of each of the number of grades assigned to the four luminance signal blocks in the macroblock. [4] The adaptive quantization apparatus of claim 3, wherein the levels of the macroblocks are classified into 32 levels based on a combination of four grade numbers (0, 1, 2, 3). 4. The apparatus of claim 3, wherein the macroblock is assigned a lower number as the complexity becomes lower. The adaptive quantization apparatus of claim 2, wherein the priority macroblock is a macroblock having the lowest level in a segment. The method according to claim 2, wherein the switching means outputs a quantized number to the first quantization means if the difference between the data amount and the reference value is less than or equal to a predetermined value, and if the difference between the data amount and the reference value is greater than or equal to the predetermined value, the encoded means is encoded. Adaptive quantization device considering the image complexity, characterized in that for outputting data to the control circuit. An adaptive quantization method of a video encoding apparatus, comprising: a first step of determining the number of ranks of discrete cosine transformed blocks and determining the level of a macroblock according to the number of ranks; Determining a quantization number according to the level of the macroblock, and quantizing and encoding the quantization number so as to adjust the quantization number so that the amount of generated data is closest to an arbitrary reference value; And a third step of quantizing using the determined quantization number. 9. The method of claim 8, wherein the second step comprises: (a) reading an quantization number table of a previous frame to determine an initial quantization number of a priority macroblock in a segment; (B) determining each of the quantization numbers of the remaining blocks by adding or subtracting them in proportion to the difference between the levels based on the quantization numbers of the priority macro blocks; (C) performing quantization and encoding by using the determined quantization number; (D) comparing the encoded data amount with an arbitrary reference value to determine whether it is closest to the reference value; (E) outputting a quantized number when it is closest in the determination; And (f) adjusting the quantization number according to the difference between the data amount and the reference value when it is not the closest in the determination. 10. The method of claim 9, wherein in step (f), if the difference between the data amount and the reference value is greater than a predetermined value, the quantization number of the remaining macro blocks in the segment is controlled by adjusting the quantization number of the priority macroblock. Adaptive quantization method considering image complexity. 10. The method of claim 9, wherein in step (f), when the difference between the data amount and the reference value is smaller than a predetermined value, if the data amount is smaller than the reference value, the quantization number is subtracted by one step from the priority macroblock and the generated data amount is subtracted. Adaptive quantization method considering the image complexity, characterized in that for adjusting the number of quantization of each macroblock compared to the reference value. 10. The method of claim 9, wherein in step (f), when the difference between the data amount and the reference value is smaller than a predetermined value, if the data amount is larger than the reference value, the quantization number is added by one step from the priority macroblock and the generated data amount is added to the reference value. Adaptive quantization method considering image complexity, characterized in that the quantization number of each macroblock is adjusted compared to a reference value.

Images