KR100230277B1 - One dimension signal adaptive filter and filtering method for reducing block effect - Google Patents

Abstract

The present invention relates to a one-dimensional signal adaptation filtering method and a one-dimensional signal adaptation filter. When one frame is composed of blocks having a predetermined size, the one-dimensional signal adaptation filtering method for reducing the blocking effect of the image data includes a predetermined quantization step ( Q) calculating a threshold value T by a function; An inclination calculation step of applying a one-dimensional window having a predetermined size along a boundary line of the block to perform a predetermined inclination operation on each pixel constituting the one-dimensional window; A binary edge map generation step of comparing the result value of each pixel in the calculated one-dimensional window with the calculated threshold value T and generating the result as a binary value for each pixel; A weighting value generation step of applying weighted values by binary edge map information belonging to a one-dimensional filter window by applying a one-dimensional filter window of a predetermined size to the generated binary edge map information; And generating a new pixel value by filtering using the generated weight value.

According to the present invention, by removing the block noise from the block-based image, the effect of improving the quality of the compressed and restored image.

Description

One-Dimensional Signal Adaptation Filter and Filtering Method for Reducing Blocking Effect

The present invention relates to data filtering, and more particularly, to a one-dimensional signal adaptation filter and a one-dimensional signal adaptation filtering method for reducing block noise.

Most picture coding standards, including MPEG by the International Organization for Standardization (ISO) and H.263 by the International Telecommunication Union (ITU), are generally block-based motion estimation and block (block) Use Discrete Consine Transform (DCT) processing. The block-based encoding causes a well-known blocking effect, especially when the image is compressed. Typical blocking effects include grid noise in homogeneous areas with relatively similar pixel values between adjacent pixels, and staircase in which image edges appear as steps along the edges of the image. noise).

When the compressed data is reconstructed and displayed on the screen, the trellis noise causes the track processed based on the block to appear at the edges between the blocks so that the viewer can know the edges between the blocks. In addition, the stair noise likewise causes the edges of the image to appear in a staircase shape, causing the viewer to feel the edges of the image uneven.

Several methods have been proposed to reduce the block effect that occurs when encoding based on such a block. First, in H.261, the blocking effect is reduced by using a simple 3 × 3 lowpass filter as a loop filter. However, when the 3x3 low pass filter is used, the computational amount increases and there is a limit in removing the blocking effect.

The present invention has been made to solve the above-described problems, and by using a one-dimensional filter window and using a one-dimensional signal adaptation filter to greatly reduce the blocking effect generated when encoding on the basis of the block, the blocking effect of the high compression encoding system. It is an object of the present invention to provide a one-dimensional signal adaptation filtering method and a one-dimensional signal adaptation filter for reduction.

1 is a block diagram showing the configuration of an embodiment of a one-dimensional signal adaptation filter according to the present invention.

2 illustrates a block and a one-dimensional window when an image frame is divided into blocks having an 8x8 (or 16x16) pixel size.

3 is a flowchart illustrating the operation of one embodiment of the present invention.

4A and 4B show basic weighting values of the filter window applied to the pixels s1 and s2.

5A to 5C illustrate weighting values and pixel values according to pixel information of the one-dimensional filter window.

6A to 6E illustrate pixel weights and pixel values for integer operations of the one-dimensional filter window.

<Description of the symbols for the main parts of the drawings>

100: image storage unit, 110: threshold calculator, 120: gradient calculator

130: threshold comparison unit, 140: binary edge map generation unit

150: filter weight generation unit, 160: one-dimensional weight filter

According to the present invention for achieving the above object, the one-dimensional signal adaptive filtering method for reducing the blocking effect of the image data when one frame is composed of blocks having a predetermined size, by a predetermined quantization step (Q) function Calculating a threshold value T; An inclination calculation step of applying a one-dimensional window having a predetermined size along the boundary line of the block to perform a predetermined inclination operation on each pixel constituting the one-dimensional window; A binary edge map generation step of comparing a result value of each pixel in the gradient one-dimensional window with the calculated threshold value T and generating the result as a binary value for each pixel; A weighting value generation step of applying weighted values by binary edge map information belonging to the one-dimensional filter window by applying a one-dimensional filter window of a predetermined size to the generated binary edge map information; And generating a new pixel value by filtering using the generated weight value.

The one-dimensional window of the gradient operation step includes a one-dimensional horizontal window whose center pixels are positioned with the boundary line of the block interposed therebetween and having a size of 1 × 4; And a one-dimensional vertical window having a size of 4 × 1.

If the binary value to be generated for each pixel in the binary edge map generation step is different for the horizontal and vertical windows, the edge is determined.

If the quantization step Q of the quantizer is smaller than the predetermined value N1, no filtering is performed, and if the frame to be filtered is an intraframe, the threshold value T is calculated. Is set to 2Q-4 for pixels adjacent to the boundary and Q + 2 for other pixels. If the frame to be filtered is an interframe and Q is greater than N1 and less than a predetermined value N2, a threshold value T is set to Q, and the frame to be filtered is an interframe and the Q Is greater than or equal to N2, the threshold value T is preferably set to a predetermined value N3.

The inclination calculation for each pixel in the one-dimensional window, which is performed in the inclination operation step, is characterized in that the absolute value of the difference between each pixel and the adjacent pixel in the one-dimensional window.

로 계산되고, 상기 픽셀 p1에 대해서는

로, 상기 픽셀 p2는

로, 상기 픽셀 p3는

로 계산되고, 4 x 1 의 크기를 갖는 일차원 수직윈도우에 대해서도 상기 수평윈도우와 같은 원리로 상기 수직윈도우내의 각 픽셀에 대한 연산이 계산됨을 특징으로 한다.The four pixels in the one-dimensional horizontal window having the size of 1 x 4 are called p0, p1, p2, and p3 from the left and the corresponding pixel values are a, b, c, d, and the pixel to the right of the pixel p3 is p4. When the corresponding pixel value is e, the gradient operation value for the pixel p0 is

For the pixel p1

The pixel p2 is

Where the pixel p3 is

Computation is calculated for the one-dimensional vertical window having a size of 4 x 1, the operation for each pixel in the vertical window is calculated on the same principle as the horizontal window.

The one-dimensional filter window of the weight generation step is a filter window having a size of 1 × 4. When the four pixels included in the one-dimensional filter window are s0, s1, s2, and s3 from the left side, the one-dimensional filter window of the weight generation step The filter window is applied only to the pixel s1 and the pixel s2, and the default weight of the filter window applied to the pixel s1 is (1, 2, 1, 1), and the default weight of the filter window applied to the pixel s2. Is (1, 1, 2, 1). The weighting value of the filter window for the pixel s1 does not generate a weighting value if the pixel s1 is an edge, and if the pixel s1 is not an edge and other pixels s0, s2, s3 in the filter window are not edges as well, the default weighting value (1, 2, 1, 1), and if the other pixels (s0, s2, s3) in the filter window are edges, the weight of the pixel that is an edge with respect to the basic weight is set to 0, but the weight of pixel s2 is If it is set to 0, the weight of pixel s3 is also set to 0. Weight generation of the filter window for the pixel s2 is also performed in the same principle as the pixel s1.

When four pixels included in the one-dimensional filter window are s0, s1, s2, and s3 from the left, the one-dimensional filter window of the weighting step is applied only to the pixels s1 and s2, and the weighting step and the pixel The value generating step is characterized in that a new pixel value is generated by applying a preset weighting value according to the edge information for each pixel of the one-dimensional filter window.

According to the present invention for achieving the above object, the one-dimensional signal adaptation filter for reducing the image blocking effect, the image storage unit for temporarily storing the image data; A threshold calculator which calculates a threshold value T by a predetermined quantization step function of the image storage unit; When the image frame of the image storage unit is divided into blocks having a predetermined size, a predetermined gradient operation is performed on each pixel constituting the one-dimensional window by applying a one-dimensional window having a predetermined size along the boundary line of the block. Gradient calculation unit to be; A threshold comparison unit for comparing a result value for each pixel of the one-dimensional window calculated by the gradient calculator with a threshold value T calculated by the threshold calculator; A binary edge map generation unit generating a result of the comparison in the threshold comparing unit as a binary value for each pixel; A filter weighting value generation unit generating weighted values by binary edge map information belonging to the one-dimensional filter window by applying a one-dimensional filter window of a predetermined size to the binary edge map information generated by the binary edge map generation unit; And a one-dimensional weighting filter for generating new pixel values by filtering using the weighting values generated by the filter weighting value generator.

Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings. FIG. 1 is a block diagram showing the configuration of an embodiment of a one-dimensional signal adaptation filter according to the present invention. The image storage unit 100, the threshold calculator 110, the gradient calculator 120, and the threshold comparator 130 are shown in FIG. ), A binary edge map generator 140, a filter weight value generator 150, and a one-dimensional weight filter 160. 3 is a flowchart illustrating the operation of the embodiment of the present invention.

The image storage unit 100 temporarily stores image data including a block effect.

The threshold calculator 110 receives a predetermined quantization step Q from the image storage unit 100 and calculates a threshold value T. In the present embodiment, if the quantization step Q of the quantizer is less than 4, no filtering is performed (step 300).

The threshold value T determines whether the frame to be filtered is an intraframe (step 305).

에 대해서는 2Q - 4로 설정하고, 그 외의 픽셀들(p0, p2, p3)에 대해서는 Q + 2로 설정한다.(310단계)The four pixels in the one-dimensional horizontal window 210 having the size of 1 x 4 are called p0, p1, p2, and p3 from the left and the corresponding pixel values are a, b, c, d, and the right of the pixel p3. When the pixel is p4 and the corresponding pixel value is e, if the frame to be filtered is an intra frame, the threshold value T is calculated for the slope of the pixel p1 adjacent to the block boundary line.

For example, 2Q-4 is set, and Q + 2 is set for the other pixels p0, p2, and p3 (step 310).

If the frame to be filtered is an interframe and the Q is less than 19 (step 315), and if it is less than 19, the threshold value T is set to Q (step 320). If Q is equal to or greater than 19, the threshold value T is set to 19 (step 325).

When the image frame of the image storage unit 100 is divided into blocks having a predetermined size, the gradient calculator 120 has a one-dimensional window having a predetermined size, preferably 1 × 4, along the boundary line of the block. In operation 330, when the image frame is divided into blocks having an 8x8 (or 16x16) pixel size, each pixel constituting the one-dimensional window is applied. The block 200 and one-dimensional windows 210 and 220 are illustrated. The one-dimensional windows 210 and 220 are positioned with their center pixels b and c interposed between the boundary line 230 of the block 200 and the one-dimensional horizontal window 210 having a size of 1 × 4. It consists of a one-dimensional vertical window 220 having a size of 4 x 1.

로 계산되고, 상기 픽셀 p1에 대해서는

로, 상기 픽셀 p2는

로, 상기 픽셀 p3는

로 계산된다. 마찬가지로 상기 4 x 1 의 크기를 갖는 일차원 수직윈도우(220)에 대해서도 상기 수평윈도우(210)와 같은 원리로 상기 수직윈도우(220)내의 각 픽셀에 대한 연산이 계산된다. 여기서, 상기 경사도연산은 상기 블록(200)의 좌측경계선(230) 및 위쪽경계선(240)에 대해서만 수행하면 된다. 상기 블록(200)의 우측경계선(250) 및 아래쪽경계선(280)은 다음 블록(270) 및 아래쪽블록(280)에서 처리된다.The gradient operation for each pixel in the one-dimensional window is calculated by taking an absolute value of the difference between the pixel and an adjacent pixel. That is, the slope calculation value for the pixel p0 is

For the pixel p1

The pixel p2 is

Where the pixel p3 is

Is calculated. Similarly, for the one-dimensional vertical window 220 having the size of 4 × 1, the calculation for each pixel in the vertical window 220 is calculated on the same principle as the horizontal window 210. Here, the gradient operation may be performed only on the left boundary line 230 and the upper boundary line 240 of the block 200. The right boundary line 250 and the lower boundary line 280 of the block 200 are processed in the next block 270 and the lower block 280.

The threshold comparison unit 130 compares the result value of each pixel of the one-dimensional window calculated by the gradient calculator 120 with the threshold value T calculated by the threshold calculator 110 to determine whether it is an edge. In operation 335, the binary edge map generator 140 generates the result of the comparison in the threshold comparator 130 as a binary value for each pixel (step 340).

Meanwhile, the threshold comparison unit 130 and the binary edge map generation unit 140 may be implemented as a module when implemented in software, and the module may be referred to as a binary edge map information generation unit 170. It's okay.

The operation of the threshold comparing unit 130 and the easy edge map generation unit 140 will be described in more detail as follows. As a result of comparing the gradient calculation value calculated for each pixel by the threshold comparison unit 130 with the threshold value T calculated by the threshold calculator 110, the gradient calculation result value for each pixel is determined by the threshold value ( If larger than T), it is determined as an edge, and the edge information edge [0] = 1 of the first pixel p0 is set. If it is smaller than the threshold value T, it is determined that it is not an edge and the edge [0] = 0 is set. Similarly, binary edge map information is obtained for edge [1], edge [2] and edge [3]. In this way, binary edge map information is generated in the horizontal and vertical directions by applying a one-dimensional window along the boundary line of the block. On the other hand, by applying the one-dimensional window in both the horizontal and vertical directions, the pixel whose binary edge information is calculated differently for both horizontal and vertical directions is determined as an edge, and the binary edge information is set to 1.

The filter weight value generation unit 150 applies a one-dimensional filter window having a size of 1 × 4 to the binary edge map information generated by the binary edge map generation unit 140 to apply a binary edge map belonging to the one-dimensional filter window. Generate weights by information. The size of the one-dimensional filter window is not limited to 1 x 4 may be further extended, it will be apparent to those skilled in the art that the present invention is not limited.

The one-dimensional weighting filter 160 filters the data of the image storage unit 100 using the weighting values generated by the filter weighting value generation unit 150 to generate new pixel values. When the four pixels in the one-dimensional filter window are s0, s1, s2, s3 from the left and the corresponding filter coefficients (or weights) are w1, w2, w3, w4, the filtering is performed. Applies only to s1 and pixel s2. That is, by performing the one-dimensional weighting filter 160 by using the weighting value generated by the filter weighting value generation unit 150, it is newly generated as pixel values for the pixels s1 and s2. That is, if the pixel values of the pixels s1 and s2 are a2 and a3 in the input image storage unit 100, the signal adaptive filtering is performed through the filter weight value generation unit 150 and the one-dimensional weight filter 160. is changed to a'3.

The operation of the filter weight value generation unit 150 and the one-dimensional weight filter 160 will be described in more detail. First, the default weight of the filter window applied to the pixel s1 is set to (1, 2, 1, 1) as shown in FIG. 4A, and the default weight of the filter window applied to the pixel s2 is shown in FIG. 4B. Set to (1, 1, 2, 1) as shown.

Then, to determine the weight value of the filter window for the pixel s1, it is checked whether the pixel s1 is an edge (step 345), and if it is an edge, one-dimensional filtering is not performed without generating a weight value (step 360). If the pixel s1 is not an edge, an appropriate weighting value is set according to the values of other pixels s0, s2, and s3 in the filter window (step 350). More specifically, for example, the pixel s1 is not an edge and the filter is set. If other pixels S0, S2, S3 in the window are not edges, they are determined by the filter weighting value shown in FIG. 4A and filtered. If the pixel s1 is not an edge and there are edges at other pixels s0, s2, and s3 in the filter window, if the pixel s0 is an edge by examining the pixels s0 and s1, the corresponding weight value w2 is shown in FIG. ) Is set to 0, and if the pixel s2 is an edge, the weight values w3 and w4 corresponding to the pixels s2 and s3 are set to 0 as shown in FIG. 5B. If pixel s3 is an edge, set the weighting value w4 corresponding to pixel s3 to zero as shown in FIG. 5C. Similarly, the weight of the one-dimensional filter window is set for the pixel s2 in the same principle as the pixel s1.

If the weighting value is set as described above, weighting filtering is performed (step 355). An example of weighting filtering according to the set weighting value is as follows. When the original pixel values of the one-dimensional filter window are a, b, c, and d, the pixel value a'2 of pixel s1 is (2b + c + d) / 4 + 0.5 with respect to the weighting value shown in FIG. 5A. It is calculated to be the value obtained by taking only the integer part. 'Int' in FIGS. 5A to 5C means that only integer values are taken. Similarly, a'2 for FIG. 5B takes an integer value by calculating (a + 2b) / 3 + 0.5, and in FIG. 5C, takes an integer value by calculating (a + 2b + c) / 4 + 0.5. . Filtering is performed on the pixel s2 based on the same principle as the pixel s1.

Meanwhile, the filtering as illustrated in FIGS. 5A to 5C may take a lot of processing time in some cases because a floating point operation is performed. Therefore, it can be treated as integer calculation. Each pixel of the 1 × 4 one-dimensional pixel window is assumed to be s0, s1, s2, s3 and the corresponding pixel values are a, b, c, d. 6A shows the weighting value when the pixel s0 is an edge. Here, the hatched portion indicates an edge, and the weight of the edge pixel is zero. The pixel value a'2 = (2a + c + d) >> 2 for the pixel s1 is calculated. In other words, (2a + c + d) is calculated and shifted to the left twice. Wherein '>>' represents a right shift operation. Similarly, FIG. 6B shows weighting values when pixels s2 and s3 are edges, and is calculated as pixel value a'2 = (6a + 10b) >> 4 for pixel s1. Fig. 6C shows a weighting value when there is no pixel as an edge, and is calculated as pixel value a'2 = (a + 4b + 2c + d) >> 3 for pixel s1. FIG. 6D shows a weighting value when the pixel s3 is an edge and is calculated as pixel value a'2 = (a + 2b + c) >> 2 for the pixel s1. Fig. 6E shows weighting values when pixels s0 and s3 are edges, and is calculated as pixel value a'2 = (10b + 6c) >> 4 for pixel s1. The pixel value a'2 for the pixel s2 is also calculated on the same principle as the pixel value calculation for the pixel s1.

It will be apparent to those skilled in the art that the above-described embodiment can be used not only as a loop filter of the encoding apparatus but also as a decoding apparatus. In addition, even if the filtering is applied only to the inside of the block having a size of 8 x 8, Mosquito noise can be reduced, and a description thereof will be omitted since it will be apparent to those skilled in the art.

According to the present invention, the effect of improving the quality of a compressed restored image is obtained by removing block noise from a block-based compressed restored image.

In addition, it is possible to reduce mosquito noise even if the signal adaptation filtering is applied only to the inside of a block having a size of 8 × 8.

Claims (13)

A one-dimensional signal adaptive filtering method for reducing the blocking effect of image data when one frame is composed of blocks having a predetermined size, the method comprising: calculating a threshold value (T) by a predetermined quantization step (Q) function; An inclination calculation step of applying a one-dimensional window having a predetermined size along the boundary line of the block to perform a predetermined inclination operation on each pixel constituting the one-dimensional window; A binary edge map generation step of comparing a result value of each pixel in the gradient one-dimensional window with the calculated threshold value T and generating the result as a binary value for each pixel; A weighting value generation step of applying weighted values by binary edge map information belonging to the one-dimensional filter window by applying a one-dimensional filter window of a predetermined size to the generated binary edge map information; And generating a new pixel value by filtering by using the generated weight value. The method of claim 1, wherein the one-dimensional window of the gradient calculation step comprises: a one-dimensional horizontal window whose center pixels are positioned with the boundary line of the block interposed therebetween and having a size of 1 × 4; And a one-dimensional vertical window having a size of 4 × 1. 3. The one-dimensional signal adaptive filtering method according to claim 2, wherein in the binary edge map generation step, if a binary value to be generated for each pixel is different for horizontal and vertical windows, it is determined as an edge. The method according to claim 1 or 2, wherein if the quantization step Q of the quantizer is smaller than a predetermined value N1, no filtering is performed, and the threshold T calculation step is to be filtered. If the frame is an intraframe, the threshold T is set to 2Q-4 for the left pixel p1 and the top pixel bordering the block boundary, and for other pixels p0, p2, and p3. If it is set to Q + 2 and the frame to be filtered is an interframe and the Q is larger than the N1 and smaller than the predetermined value N2, the threshold value T is set to Q and the frame to be filtered. And if Q is equal to or greater than N2, the threshold value T is set to a predetermined value N3. The method of claim 4, wherein N1 = 4 and N2 = N3 = 19. The method of claim 1 or 2, wherein the gradient calculation for each pixel in the one-dimensional window, which is performed in the gradient calculation step, is an absolute value of a difference between pixels adjacent to each pixel in the one-dimensional window. Signal adaptive filtering method. The method of claim 6, wherein the four pixels included in the one-dimensional horizontal window having a size of 1 x 4 are p0, p1, p2, and p3 from the left side, and the corresponding pixel values are a, b, c, d, and the pixel p3. When the pixel on the right is p4 and the corresponding pixel value is e, the gradient operation value for the pixel p0 is

For the pixel p1

The pixel p2 is

Where the pixel p3 is

And a calculation for each pixel in the vertical window for the one-dimensional vertical window having a size of 4 x 1, in the same principle as the horizontal window. The one-dimensional signal adaptive filtering method of claim 1, wherein the one-dimensional filter window of the weight generation step is a filter window having a size of 1 × 4. The method of claim 8, When four pixels included in the one-dimensional filter window are s0, s1, s2, and s3 from the left, the one-dimensional filter window of the weighting step is applied only to the pixels s1 and s2, and is applied to the pixels s1. The default weight of the filter window is (1, 2, 1, 1), and the default weight of the filter window applied to the pixel s2 is (1, 1, 2, 1), The weighting value of the filter window for the pixel s1 does not generate a weighting value if the pixel s2 is an edge, and if the pixel s1 is not an edge and the other pixels s0, s2, s3 in the filter window are also edges, the default weighting value (1, 2, 1, 1), and if the other pixels (s0, s2, s3) in the filter window are edges, the weight of the pixel that is an edge with respect to the basic weight is set to 0, but the weight of pixel s2 is If it is set to 0, the weighting value of the pixel s3 is also set to 0, and the weighting value generation of the filter window for the pixel s2 is performed on the same principle as the pixel s1. The method of claim 8, wherein when the four pixels included in the one-dimensional filter window are s0, s1, s2, and s3 from the left, the one-dimensional filter window of the weighting step is applied only to the pixels s1 and s2, The weight value generation step and the pixel value generation step may be performed by applying a predetermined weight value according to edge information of each pixel of the one-dimensional filter window to perform a bit shift operation to generate a new pixel value. One-dimensional signal adaptive filtering method. An image storage unit for temporarily storing image data; A threshold calculator which calculates a threshold value T by a predetermined quantization step function of the image storage unit; When the image frame of the image storage unit is divided into blocks having a predetermined size, a predetermined gradient operation is performed on each pixel constituting the one-dimensional window by applying a one-dimensional window having a predetermined size along the boundary line of the block. Gradient calculation unit to be; A threshold comparison unit for comparing a result value for each pixel of the one-dimensional window calculated by the gradient calculator with a threshold value T calculated by the threshold calculator; A binary edge map generation unit generating a result of the comparison in the threshold comparing unit as a binary value for each pixel; A filter weighting value generation unit generating weighted values by binary edge map information belonging to the one-dimensional filter window by applying a one-dimensional filter window of a predetermined size to the binary edge map information generated by the binary edge map generation unit; And a one-dimensional weighting filter for generating a new pixel value by filtering using the weighting value generated by the filter weighting unit. 12. The method of claim 11, wherein the one-dimensional window of the gradient calculation unit comprises: a one-dimensional horizontal window whose center pixels are positioned with the boundary line of the block interposed therebetween and having a size of 1 × 4; And a one-dimensional vertical window having a size of 4 × 1. The one-dimensional signal adaptation filter according to claim 11, wherein the one-dimensional weighting filter has a size of 1 × 4.

Images