KR100711025B1 - The method for filtering a residual signal to improve performance in the standard coding mode of motion picture - Google Patents

Abstract

The present invention relates to a residual signal filtering method for improving the performance of a video standard coding scheme. In particular, the present invention relates to quantization error based on H.264 and to correlations between pixels of adjacent blocks and current pixels. The present invention relates to a method for removing noise from an input image and filtering a bit rate problem by filtering.

Residual signal filtering method for improving the performance of the video standard coding method of the present invention comprises the steps of: a residual signal preprocessing filter to obtain an error signal between the input image and the motion-compensated image or the input image and the intra-predicted image; A second step of removing the high frequency component by low frequency filtering the error signal by the residual signal preprocessing filter; A third step of performing a discrete cosine transform (DCT) on the low frequency filtered error signal; A fourth step of quantizing the DCT error signal; Obtaining noise in the quantized error signal; And a sixth step of obtaining an average energy of the noise.

Video, Encoding, Residual Signal, Filtering

Description

The method for filtering a residual signal to improve performance in the standard coding mode of motion picture}

1 shows a filtering method using a residual signal preprocessing filter in an encoder of the present invention.

Figure 2 shows the form of the one-dimensional filter of the present invention.

The present invention relates to a residual signal filtering method for improving the performance of a video standard coding scheme. In particular, the present invention relates to quantization error based on H.264 and to correlations between pixels of adjacent blocks and current pixels. The present invention relates to a method for removing noise from an input image and filtering a bit rate problem by filtering.

The JVT method, which has been developed as the next generation video compression method since 1997, has proved superior in performance compared to the conventional H.263 and MPEG4 methods. The main difference between the JVT method and the H.263 and MPEG4 methods is that the JVT method uses 4 × 4 block-based change and encoding, transform block size motion estimation, compensation, and one variable length code (VLC). Is in point.

H.264 features an integer DCT transform with a 4 × 4 block size, a motion vector prediction with variable block size that requires a lot of computation when encoding, intra coding with various directions, and multiple reference frame motion estimation. I am doing it. In image encoding, the purpose of the information is not to preserve the information but to approximate the information. Therefore, the reconstructed image is accompanied by some distortion.

However, the conventional H.264 does not provide a method for removing noise and visually inconvenient quantization noise and reducing a transmission bit rate generated in an image acquisition process from an original image in an environment having a limited channel bandwidth and a band band.

Accordingly, the present invention is to solve the above problems, by using the correlation between the quantization error and the pixel of the adjacent block and the current pixel based on H.264 to remove the noise generated from the input image by filtering the pixel of the current block And to provide a solution to the bit rate problem.

The present invention relates to a residual signal filtering method for improving the performance of a video standard coding scheme. In particular, the present invention relates to a pixel of a current block using a correlation between a quantization error and a pixel of an adjacent block and a current pixel based on H.264. The present invention relates to a method for filtering noise and removing noise from an input image and solving a bit rate problem.

In the residual signal filtering method for improving the performance of the video standard coding method of the present invention, a residual pre-filter obtains an error signal between an input image and a motion compensated image or an input image and an intra predicted image. Stage 1; A second step of removing the high frequency component by low frequency filtering the error signal by the residual signal preprocessing filter; A third step of performing a discrete cosine transform (DCT) on the low frequency filtered error signal; A fourth step of quantizing the DCT error signal; Obtaining noise in the quantized error signal; And a sixth step of obtaining an average energy of the noise.

Hereinafter, with reference to the accompanying drawings, the configuration and operation of the embodiment of the present invention will be described in detail.

The video compression method transmits temporally and spatially compressed information and additional information necessary for decoding by using a method of removing spatial and temporal excess information from the encoder, and the decoder is configured in the reverse order of the encoder.

In the present invention, a residual signal filter method for filtering a current block from information of a quantization error and an adjacent block will be described.

The residual signal preprocessing filter effectively removes a high frequency component through a low frequency filtering process to remove an error signal (DFD: Displaced Frame Difference (DFD)) between an input image and a motion compensated image or an input image and an intra prediction image. After the quantization process, it is used to improve the compression efficiency in the run-length process. In the filtering process, since the error error distribution that is not explained differs according to encoding characteristics of each macro block, three macro blocks are classified and filtered as follows.

First, 16 × 16 inter MB

Second, 16 × 16 Intra MB

Third, 4 × 4 Intra MB

When the encoding form of the adjacent previous macro block and the current macro block is different according to the above macro blocks, the adjacent pixels are not used in the filtering process.

1 shows a filtering method using a residual signal preprocessing filter in an encoder of the present invention. In general, image information of the input video is compressed through entropy coding (200) after going through a discrete cosine transform (DCT) and a quantization (100) in the case of an intra frame, and in the case of an inter frame thereafter. The compressed image is compressed based on the difference between the decoded image information and the newly input image information. The conventional compressed image is subjected to inverse quantization and inverse discrete cosine transform (IDCT) 300 in the reverse order of the compression method. The processed image is decoded or encoded through intra prediction (400), motion compensation (500), and motion estimation (600), and processed to block blocking between pixels. The deblocking process is used.

In the present invention, the residual signal preprocessing filter 800 is positioned directly in front of the DCT to filter the intensity value in the intra block and to filter the DFD, that is, the residual signal in the inter block, to improve the image quality. .

Although the acquired image basically has a degree of difference depending on the image capturing apparatus, the attached noise exists and the degradation of the image quality is more serious due to the quantization process. In other words, as the quantization step size increases, the quantization error value increases, resulting in loss of high frequency components. In the filtering process performed in the present invention, a filter has a one-dimensional Gaussian model such that the characteristic of the filter is represented under Equation 1 under the assumption that the statistical distribution of the image has a Gaussian distribution.

영상의 국부지역 분산으로 고주파 영역일수록 큰 값을 가지며, 이는 가우시안 필터링 영상의 열화정도를 감소시켜 에지정보를 보존한다. k는 스무딩(smoothing)을 결정하는 매개변수로서 국부지역 평균값을 이용하여 나타내었다.

는 양자화로 인한 노이즈의 분산으로 H.264 부호화부에서 발생하는 양자화 노이즈를 제거하기 위한 특성을 반영한다. 양자화로 인한 노이즈를 결정하기 위하여 다음과 같은 수학식을 기반으로 노이즈 정도를 유추하였다.In Equation 1

The local region of the image is distributed, and the higher frequency region is, the larger the value is. k is expressed using the local mean value as a parameter for determining smoothing.

Reflects a characteristic for removing quantization noise generated in the H.264 encoder due to variance of noise due to quantization. In order to determine noise due to quantization, the noise degree was inferred based on the following equation.

In the H.264 video coding method, DCT and quantization are performed together. The H.264 DCT and quantization process is as follows.

는 성분 곱(element product)이다.In Equation 2, X is an input 4 X 4 matrix, Y is a transformed matrix, Z is a transform and quantized matrix, QE is a quantization matrix including a scale part of DCT in the H.264 video standard coding method, and qp is Quantization step size,

Is an element product.

A typical 4 x 4 DCT process is as follows.

only,

Matrix A in Equation 3 may be separated as in Equation 4.

Equation 4 may be rewritten as Equation 5 below.

In Equation 5 using Equation 4, Y represents an integer conversion method of H.264. The quantization process of Equation 2 is expressed as Equation 6.

In the H.264 video coding method, S in Equation 6 is included in a quantization process to form a new quantization matrix that performs different quantization according to pixel positions of an input matrix. In the H.264 video coding method, a quantization matrix is expressed as in Equation 7 using Equation 6.

In Equation 7, S is a scale matrix generated for DCT integer conversion, and QP is a general quantization coefficient matrix. Therefore, it may be expressed as Equation (8).

은 수학식 9와 같다.In Equation 8

Is the same as Equation 9.

The quantization process in the H.264 video coding method is as follows.

Substituting Equation 8 into Equation 10 results in Equation 11.

이므로 상기 수학식 11은 수학식 12와 같다.The transformed matrix is

Equation 11 is equal to Equation 12.

In Equation 12, the quantization error is represented by Equation 13.

In the H.264 standard coding scheme, QE is represented by Equation 14. In Equation 14, qp% 6 is the remainder obtained by dividing the quantization step size by six.

By substituting Equation 14 into Equation 13, quantization noise shown in Table 1 can be obtained.

Table 1 shows the quantization error of the present invention in the frequency domain. In the time domain, the quantization error matrix is obtained by IDCT.

qp / 6 = 0 qp / 6 = 1 qp / 6 = 2 qp / 6 = 3 qp / 6 = 4 qp / 6 = 5 qp / 6 = 6 qp / 6 = 7 qp / 6 = 8 qp% 6 = 0 MAX 0.64 1.28 2.56 5.12 10.24 20.48 40.96 81.92 163.84 MEAN 0.63 1.27 2.54 5.08 10.16 20.32 40.64 81.28 162.56 qp% 6 = 1 MAX 0.7 1.41 2.82 5.64 11.28 22.56 45.12 90.24 180.48 MEAN 0.7 1.4 2.8 5.6 11.2 22.4 44.8 89.6 179.2 qp% 6 = 2 MAX 0.81 1.63 3.26 6.52 13.04 26.08 52.16 104.32 208.64 MEAN 0.79 1.59 3.18 6.36 12.72 25.44 50.88 101.76 203.52 qp% 6 = 3 MAX 0.9 1.8 3.6 7.2 14.4 28.8 57.6 115.2 230.4 MEAN 0.89 1.78 3.56 7.12 14.24 28.48 56.96 113.92 227.84 qp% 6 = 4 MAX One 2 4 8 16 32 64 128 MEAN 0.99 1.98 3.96 7.92 15.84 31.68 63.36 126.72 qp% 6 = 5 MAX 1.14 2.28 4.56 9.12 18.24 36.48 72.96 145.92 MEAN 1.13 2.27 4.54 9.08 18.16 36.32 72.64 145.28

The QP_error_time matrix is a value obtained by IDCTing the QP_error_freq matrix and the values are collected around the DC. The coefficients corresponding to the (0,0) coefficients of the QP_error_time matrix are shown in Table 2.

qp / 6 = 0 qp / 6 = 1 qp / 6 = 2 qp / 6 = 3 qp / 6 = 4 qp / 6 = 5 qp / 6 = 6 qp / 6 = 7 qp / 6 = 8 qp% 6 = 0 2.34 4.69 9.38 18.76 37.52 75.05 150.10 300.20 600.40 qp% 6 = 1 2.57 5.13 10.26 20.53 41.05 82.11 164.21 328.42 656.85 qp% 6 = 2 2.94 5.88 11.76 23.52 47.04 94.08 188.17 376.33 752.66 qp% 6 = 3 3.29 6.57 13.14 26.28 52.57 105.13 210.27 420.53 841.06 qp% 6 = 4 3.66 7.32 14.64 29.28 58.55 117.11 234.21 468.42 qp% 6 = 5 4.19 8.38 16.77 33.53 67.07 134.13 268.26 536.52

The average energy of the QP_error_time matrix can be obtained as shown in Equation 16.

는 QP_error_time 행렬의 i, j 번째 계수이다.here,

Is the i, j th coefficient of the QP_error_time matrix.

The average energy of the quantization noise is obtained from Equation 16 as shown in Table 3.

qp / 6 = 0 qp / 6 = 1 qp / 6 = 2 qp / 6 = 3 qp / 6 = 4 qp / 6 = 5 qp / 6 = 6 qp / 6 = 7 qp / 6 = 8 qp% 6 = 0 0.40 1.60 6.43 25.70 102.81 411.24 1644.9 6579.8 26319 qp% 6 = 1 0.48 1.92 7.69 30.78 123.13 492.55 1970.2 7880.9 31523 qp% 6 = 2 0.630 2.52 10.08 40.33 161.31 645.24 2580.9 10323 41295 qp% 6 = 3 0.78 3.15 12.62 50.48 201.93 807.72 3230.9 12923 51694 qp% 6 = 4 0.97 3.90 15.62 62.51 250.04 1000.1 4000.6 16002 qp% 6 = 5 1.28 5.13 20.52 82.10 328.41 1313.6 5254.6 21018

은 양자화 노이즈 평균 에너지인 표 3에 의해 결정된다.

Is determined by Table 3, which is the average energy of quantization noise.

는 이전 블록의 화소값이고,

는 현재 블록의 화소값이다. 현재 블록이 입력되었을 때, 수직방향으로 필터링한 후에 수평방향으로 필터링한다. 현재 처리하는 화소값이

이라고 할 때, 필터의 매개변수는 다음과 같다. 상기 수학식 1로부터 인접된 이전 블록과 현재 블록의 화소값

과

를 이용한 국부 지역정보, 평균, 분산 및 스무딩 매개변수는 수학식 17 내지 수학식 19와 같은 식으로 표현된다.Figure 2 shows the form of the one-dimensional filter of the present invention.

Is the pixel value of the previous block,

Is the pixel value of the current block. When the current block is input, it filters in the vertical direction and then in the horizontal direction. The current pixel value

In this case, the filter parameters are as follows. Pixel values of adjacent previous blocks and current blocks from Equation 1

and

The local area information, the average, the variance, and the smoothing parameter using are expressed by Equations 17 to 19.

과

를 필터링하면 수학식 20과 같다.2 by using Equations 18 and 19

and

Is filtered as shown in Equation 20.

In Equation 20, p and q represent unprocessed pixel values, Q represents filtered pixel values, and H (0) and H (1) are filter coefficient values obtained by Equation (1). This gives a weight to the pixel to be processed currently, and gives a weight to the pixels adjacent to the left and right of the current pixel by using the symmetry of the filter.

It will be apparent to those skilled in the art that various changes and modifications can be made without departing from the technical spirit of the present invention through the above description. Therefore, the technical scope of the present invention should not be limited only to the contents described in the embodiments, but should be defined by the claims.

As described above, the residual signal filtering method for improving the performance of the video standard coding method according to the present invention is a method for improving the compression efficiency in a digital video apparatus using the JVT video compression method. It can be applied to the encoding.

Claims (8)

Obtaining, by the residual signal preprocessing filter, an error signal between the input image and the motion compensated image or the input image and the intra predicted image; A second step of removing the high frequency component by low frequency filtering the error signal by the residual signal preprocessing filter; A third step of performing a discrete cosine transform (DCT) on the low frequency filtered error signal; A fourth step of quantizing the DCT error signal; Obtaining noise in the quantized error signal; And And a sixth step of obtaining an average energy of the noise, wherein the filtering process includes a 16 × 16 inter macro block, a 16 × 16 intra macro block, and a 4 × 4 intra macro block. In this case, the residual signal for improving the performance of the video standard coding method, characterized in that the neighboring pixel is not used in the filtering process when the coding form of the adjacent previous macro block and the current macro block is different. Filtering method. The method according to claim 1, The low frequency filtering of the second step includes filtering an intensity value in an intra block and filtering a residual signal in an inter block. delete delete The method according to claim 1, The low frequency filtering is a residual signal filtering method for improving the performance of the video standard coding method, characterized in that the statistical distribution of the image has a Gaussian distribution. The method according to claim 5, The characteristics of the filter for low frequency filtering is

Residual signal filtering method for improving the performance of the video standard coding method, wherein Z is a normalization constant,

Is the variance for quantization noise,

Is the local variance, and k is the parameter that determines smoothing. The method according to claim 1, The low frequency filtering is performed when a current block is inputted. A first step of filtering in the vertical direction; And A second step of filtering in the horizontal direction; residual signal filtering method for improving the performance of the standard video coding method. The method according to claim 6, The parameter of the filter is

,

, And

Determined by, and the filtered pixel value is

And

Residual signal filtering method for improving the performance of the video standard coding scheme, characterized in that p ₁ is the pixel value of the previous block, q ₁ , q ₂ is the pixel value of the current block, Q is filtered Pixel value).

Images