KR0141297B1 - Nonlinear Quantization Method - Google Patents

Abstract

According to the present invention, in the nonlinear quantization method according to the present invention, the nonlinear quantization method according to the present invention can reduce information loss of low frequency components by performing nonlinear quantization on low frequency components and high frequency components including DC. According to the importance of the frequency component and the quantization coefficient applied to the low frequency component of the surrounding macroblock, the amount of the generated bitstream can be reduced by setting the slope of the quantization coefficient and the quantization coefficient of the current macroblock.

Description

Nonlinear Quantization Method

The present invention relates to a quantization technique in a video encoding apparatus using a transform coefficient, and is particularly suitable for adaptively performing non-linear quantization for low frequency components and high frequency components including DC in a change coefficient to reduce information loss of low frequency components. It relates to a nonlinear quantization method.

In general, a video encoding apparatus performs a discrete cosine transform after performing a preprocessing process on a video signal input for encoding and then converts the discrete cosine transformer (hereinafter, referred to as DCT), and converts the video signal output from the CDT to a predetermined value. A quantizer for quantization in the quantization step, a variable length coder (hereinafter, referred to as VLC) for variable length coding the quantized video signal in the quantizer, and a motion estimation and compensator are included.

In the conventional video encoding apparatus having the above-described configuration, in the DCT, in the case of the macroblock to which inter-frame prediction encoding is applied, the residual macroblock is made by subtracting the prediction macro block from the current macroblock, and the intra-frame encoding In this case, no change is made to the current macro block. DCT is a transformation that makes energy cohesion almost optimal for signals with large correlation coefficients, and has a good structure to reflect human sensitivity.

In the quantizer, on the other hand, weights are given for each coefficient by an appropriate weighting matrix considering human sensitivity (ie, visual characteristics). Subsequently, the macroblocks encoded in the frame are linearly quantized to 8 bits in the case of DC coefficients, and quantized by a uniform quantizer having no dead-zone in the case of AC coefficients. On the other hand, for macroblocks that are inter-frame coded, all DCT coefficients are quantized by a dead-zone uniform quantizer. At this time, the step size of the quantizer is adjusted as a quantization scale factor by a buffer adjustment algorithm.

1 is a block diagram of a general quantizer applied to an image encoding system using a transform coefficient.

In FIG. 1, the first quantizer 1 is for quantizing by a weighted quantization matrix set in consideration of a human visual effect by taking a DCT transform coefficient provided from a DCT block (not shown). The second quantizer 3 quantizes the quantization coefficient according to how much the bit stream generated while encoding is filled in the buffer, and then supplies the quantized coefficient to the VLS (not shown). That is, the first quantizer 1 has a quantization matrix of the example as shown in FIG. 2A, and the second quantizer 3 has a constant quantization coefficient that changes frequently with the example as shown in FIG. 2B.

However, in the conventional quantization method, when the transmitted bit stream fills the buffer more than a predetermined level, a large quantization coefficient is set and the second quantizer quantizes not only low frequency information but also high frequency information. If the size of is small, most of them will be quantized to zero, which will greatly reduce the amount of data. However, at this time, since the resolution of the DC component, which is a low frequency component, varies greatly, there is a problem in that a blocking effect occurs when looking at the reconstructed image at the reception side.

Accordingly, the present invention is to solve the problems of the prior art described above, in the video encoding apparatus using a transform coefficient, in order to reduce information loss of the low frequency component, the low frequency component including DC and the high frequency component of the transform coefficient are already encoded. The DC component quantization coefficients of the current macroblock are determined by referring to the DC component quantization coefficients of adjacent neighboring macroblocks, and adaptively transformed from the DC component to the high frequency component direction based on the full state of the output buffer and the variance of the macroblock. It is an object of the present invention to provide a nonlinear quantization method capable of performing adaptive nonlinear quantization by determining a slope of a quantization weight.

In order to achieve the above object, the present invention provides a non-linear quantization method of DCT transform coefficients in a macroblock unit of discrete cosine transform, wherein a predetermined weighting factor is set in consideration of a human visual effect. First-order quantization using a given quantization matrix: DC component quantization coefficients of the current macroblock to be quantized based on the data storage state information of the output buffer, the bit generation amount, and the average quantization coefficients of neighboring neighboring macroblocks already coded. Determining step: Determining a slope that is set so that the weight is greater as the distance away from the DC component toward the high frequency component based on the dispersion value of the current macro block and the data storage state information of the output side buffer: and the determined DC component quantization coefficients and the determined slope On the basis of the adaptive quantization factor resulting la, there is provided a non-linear quantization method comprising the steps in the second quantizing the DCT transform coefficients of the current macro block is adaptively.

1 is a block diagram of a general quantization apparatus suitable for applying a nonlinear quantization method according to the present invention and applied to an image encoding system using a transform coefficient.

2A and 2B show examples of quantization matrices and quantization coefficients in the first and second quantizers shown in FIG.

FIG. 3 is a graph of quantization coefficients. FIG. 3a is a graph according to a conventional method of setting quantization coefficients in the second quantizer shown in FIG. 1, and FIG. 3b is a graph of a quantization coefficient setting method according to the present invention. 3c is a graph of quantization coefficients according to the conventional quantization method and the quantization method of the present invention for comparison with each other.

* Explanation of symbols for main parts of the drawings

1,3: Quantifier

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

The quantization apparatus to which the nonlinear quantization method according to the present invention is applied includes a first quantizer 1 and a second quantizer 3, as shown in FIG. 1, wherein the nonlinear quantization method of the present invention The method of determining the quantization coefficient in the second quantizer 3 is different from the conventional method described above. Therefore, the following will be described based on this point.

According to the present invention, in the first quantizer 1, as described above, a DCT transform coefficient output from a DCT block (not shown) is input and quantized by a weighted quantization matrix set in consideration of human visual effects. It is supplied to the 2nd quantizer (3).

In addition, the second quantizer 3 determines the quantization coefficients in consideration of the importance (i.e., low frequency and high frequency components) of each frequency component. The DC component quantization coefficient of the current macroblock is determined based on the quantization coefficients, that is, the determination of QP _DC2 , as shown in FIG.

BF (0-100%): amount of data stored in buffer

B _A : allocated bit rate / frame

B _R : remaining bit rate / frame

N _T : Total number of blocks / frame

N _R : Number of remaining blocks / frame

QP _M : Average Quantization Coefficient

QP _T : Maximum Quantization Coefficient

That is, in Equation 1, QP _DC2 denotes a coefficient for quantizing a DC component, and the quantization coefficient value is influenced by the full state of the output buffer, the bit generation amount, and the average quantization coefficient value of the already encoded neighboring macroblock. Will receive.

At this time, in Equation 1, QP _DC2 is determined to be the smaller of the maximum quantization coefficient QP _T and the change amount QP _M + D _F based on the quantization coefficient of the block immediately before. Here, the change amount DF is divided into three cases, as shown in Equation 1 above. If the DD value is 0, QP _M is used as it is. This means that it generates a positive DF, increasing the quantization coefficient. In addition, if the DD value is less than 0, the quantization coefficient is slightly decreased in the reverse direction.

Here, the reason why the change width of the DF value when the plus sign is larger than the change width of the DF value when the plus sign is larger according to the sign of the DD value is to make the bit generation stronger. Therefore, when the DD value is a negative sign, the maximum change width of the quantization coefficient is 4, and when the DD value is a positive sign, the maximum change width of the quantization coefficient may be up to the maximum value QP _T depending on the amount of data stored in the buffer. have.

On the other hand, the second quantizer 3 determines the DC component quantization coefficient of the current macro block as described above, and then the variance value of the block and the output buffer determine the quantization weighting factor based on the data fullness state information, that is, from the DC component. The slope S as shown in FIG. 3B, which is a quantization weighting factor to a high frequency component, is calculated as in the following equation.

That is, the slope is determined so that the quantization weight gradually increases as the distance from the DC component increases.

DV: variance of a block

DV _T : Maximum Permissible Brirock Dispersion

S _T : Maximum slope

W (DV, B _F ): Gradient Weight

In other words. As can be seen from Equation 2, the slope of the quantization coefficient weighting factor can vary from 0 to the maximum slope S _T , where the factors affecting the change are the variance of the current macroblock and the output buffer. The data is full.

FIG. 3A is a graph for explaining a conventional quantization coefficient setting method in the second quantizer 3 shown in FIG. 1, and FIG. 3B is a graph for explaining the quantization coefficient setting method according to the present invention. That is, as can be seen from the above drawings, in the related art, a constant weighting element, for example, '1' has been given regardless of the distance to the DC component, but in the present invention, the higher frequency component becomes farther from the DC component. As the weighting factor value increases, the quantization width of the transform coefficient increases. This means that the quantization coefficients are not uniformly applied to all transform coefficients as in the above-described prior art, but are treated differently according to the importance of frequency components.

FIG. 3C shows the quantization coefficients set by the conventional method shown in FIG. 3A and the quantization coefficients set by the present invention shown in FIG. 3B on the same graph for comparison with each other, where QP ₂ is a DC and a low frequency component. in increasing the amount of the bit stream to have a greater than the conventional QP ₁ reduces the amount of bit stream generated by the high-frequency components, low-frequency component by using the small quantization step, it has a value less than the conventional QP _1, the high-frequency component by Try to maintain the accuracy of the conversion coefficient value as much as possible.

Here, QP _DC2 considers the QP _DC2 of each macroblock that is encoded with the quantization coefficient value given to the DC component of QP ₂ so that the difference value is within a predetermined range (T _DC ) so that the blocking effect does not greatly occur. Do not

As described above, according to the nonlinear quantization method of the present invention, information loss of low frequency components can be reduced by performing nonlinear quantization on low frequency components including DC in a transform coefficient. In addition, the present invention sets the slope of the quantization coefficient and the quantization coefficient of the current macroblock according to the importance of each frequency component and the quantization coefficient applied to the low frequency component of the surrounding macroblock, thereby suppressing the amount of generated bits as well as the human visual characteristics. Subjective picture quality improvement effect can be obtained.

Claims (2)

A method of nonlinear quantization of DCT transform coefficients in discrete cosine transformed macroblock units, wherein the DCT transform coefficients in macroblock units are first quantized by a quantization matrix to which a predetermined weighting factor is assigned in consideration of a human visual effect. process; Determining the DC component quantization coefficients of the current macroblock to be quantized based on the data storage state information of the output buffer, the bit generation amount, and the average quantization coefficients of neighboring neighboring macroblocks already encoded; Determining a slope that is set to be greater as the distance from the DC component toward the high frequency component based on the dispersion value of the current macro block and the data electric field state information of the output buffer; And adaptively secondary quantizing the DCT transform coefficients of the current macroblock based on the determined DC component quantization coefficients and the adaptive quantization coefficients determined according to the determined slope. The nonlinear quantization method of claim 1, wherein the determined adaptive quantization coefficient is set to a value such that a difference value between the quantization coefficient applied to low frequency components of the neighboring macroblock does not deviate from a predetermined range.