KR20050096409A - Method for detecting and compensating corner outlier - Google Patents

Abstract

The present invention discloses a method for removing edge noise among noises that cause blocking of images generated by block-based image coding. The edge noise elimination method compensates for the pixel having a very large difference with the surrounding pixel values at the intersection of the block boundary so as to be similar to the surrounding pixel values to reproduce the image with the best image quality. However, in the conventional edge noise compensation method, when an error occurs in the compensation step because the compensation area is limited to one block, there is a high possibility that the error may be diffused in the next compensation step of the adjacent pixel.

According to the present invention, it is possible to smoothly transmit a low bit rate by reducing corner noise adaptively to an image to reduce unnatural noise of a transmitted image.

Description

Method for detecting and compensating corner outlier

The present invention relates to a method for removing block noise of an image generated during encoding of blocks encoded in a block unit. Particularly, an edge component of a reconstructed image is clearly visible by effectively removing edge noise among inter-block boundary noises. An edge noise detection and compensation method is provided.

In general, most picture coding standards, including MPEG of the International Standardization Organization (ISO) and H.263 of the International Telecommunication Union (ITU), are block-based motion estimation and block (block) Use Discrete Cosine Transform (DCT) processing.

Block DCT (hereinafter referred to as 'BDCT') based coding techniques such as the JPEC method have been successfully applied in the compression of still and moving pictures. In BDCT-based compression and decompression technology, one image is divided into a plurality of non-overlapping blocks, each of which is N × N in size. In this encoding process, original image information of an N × N image block is converted by DCT. The DCT coefficients are then quantized. The quantized block image data is encoded in a form suitable for transmission through a channel, and then encoded in a form suitable for transmission on the channel or recorded in the medium, and recorded on the medium. In the DCT-based decoding process, data received through the channel or read from the medium is first decoded and then reconstructed to the original image through inverse quantization and inverse DCT.

The block-based encoding causes a blocking effect, as is well known, especially when the image is compressed. A typical blocking phenomenon is an edge of an image due to an amplification difference between grid noise in a homogeneous area where pixel values are relatively similar between adjacent pixels and boundary pixel values of neighboring blocks. There is staircase noise along the edge where the edges of the image appear as steps.

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 edge of the image to appear as a staircase, causing the viewer to feel that the edge of the image is rugged.

Several methods have been proposed to remove the blocking noise generated in the encoding process of the images encoded in such a block unit. First, in H.261, the blocking phenomenon is reduced by using a simple 3 × 3 low pass filter (LPF) as a loop filter. In addition, a simple edge loop filter has been proposed to reduce blocking and mosquito noise, and a nonlinear filter using a binary index has been proposed.

However, with these methods, smoother images can be produced by removing high frequency components, but the edges of the original images tend to appear blurry, and the compensation region crosses one block boundary. It is limited to a single block at a point, and when an error occurs in the compensation step, there is a problem that the error can be spread in the compensation step of adjacent pixels.

Accordingly, the present invention has been made to solve the above-described conventional problems, an object of the present invention is to detect the edge noise of the edge noise of the inter-block boundary noise generated in the encoding process of the image coded in units of blocks and To provide a compensation method.

Another object of the present invention is to compensate for a pixel having a very large difference from surrounding pixel values at the point where the block boundary intersects to be similar to the surrounding pixel values to obtain an image having the best image quality without discontinuity between the internal pixels of the block. To provide a method for detecting and compensating edge noise to reproduce.

It is still another object of the present invention to provide an edge noise detection and compensation method for reducing unnatural noise of a transmitted image to facilitate low bit rate image transmission.

In order to achieve the above object, the edge noise detection method according to the present invention is a method for removing edge noise of an image generated during encoding of an image encoded in a block unit. Calculating an average value and a complexity of the target block and adjacent blocks vertically and horizontally adjacent to the block; And comparing the calculated inter-block difference and complexity of the average value with a boundary value obtained by using a quantization parameter, and detecting a block in which edge noise occurs, wherein the edge noise detection step includes: (a) block boundary crossing; Comparing whether the average difference between the blocks vertically and horizontally adjacent to the point is greater than the average value, and (b) Detecting whether the area is smooth around the block boundary intersection point, (a), (b When a block that satisfies all of the steps) is detected, the block is defined as a block in which edge noise occurs.

In the step (a), after calculating the average value of several pixels representing each block, the difference between the average value and the adjacent block is calculated to detect a case where the difference value exceeds the block discontinuous boundary value (TH1), The threshold value TH1 is characterized by being calculated by 2 * QP (quantization parameter).

In the step (b), the case where the complexity obtained through the difference of pixel values in the detection object block is smaller than the complexity threshold value TH2 is detected, and the threshold value TH2 is 2 * QP (quantization parameter) + 20. It is characterized by being calculated as.

In addition, the edge noise compensation method according to the present invention, in the method for removing the edge noise of the image generated during the encoding process of the image coded in units of blocks, (c) the edge noise is mainly a block of the image border in the diagonal direction Since the position is generated at the boundary crossing point, the boundary of the image is smoothly connected using the pixel value of the adjacent block.

In the step (c), since corner noise is mainly generated in a diagonal 45 degree direction in the case of the A block, (c1) A ₁ pixel has B _1, B _3, C _{1, and} C ₂ of the B and C blocks having adjacent block boundaries. Compensating by the following equation to have a value similar to that of the pixels; A ₁ '= (2 * A ₁ + 2 * B ₁ + 2 * C ₁ + B ₃ + C ₂ + 4) / 8 (c2) A ₂ pixels are C ₂ and C ₅ pixels of C block with adjacent block boundaries. Compensating by the following equation to have a value similar to the following; And A ₂ '= (2 * A ₂ + C ₂ + C ₅ + 2) / 4 (c3) A ₃ pixels can be expressed in the equation below so that the block boundaries are similar to those of B ₃ and B ₆ pixels in adjacent B blocks. Compensating by; A ₃ ′ = (2 * A ₃ + B ₃ + B ₆ + 2) / 4.

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

4 is a block diagram of a decoding apparatus according to an embodiment of the post-processing technique of the present invention, which includes a demultiplexer 100, a formal decoder 102, a motion decoder 104, a texture decoder 106, and a motion. The compensator 108, the VOP reconstruction unit 110, the VOP memory 112, and the edge noise canceller 114 are configured.

The encoded bit string is input to the demultiplexer 100 and separated into data to be input to the shaping decoder 102, the motion decoder 104, and the texture decoder 106. After that, the shape decoder 102 decodes the feature of the input image, the motion decoder 104 decodes the motion vector value estimated by the encoder, and the texture decoder 106 determines the reference image indicated by the motion vector. And decode the difference image of the encoded original image.

The motion compensator 108 obtains the reference image indicated by the motion vector from the VOP memory 112 and sends it to the VOP reconstruction unit 110, and the VOP reconstruction unit 110 displays the characteristics of the image obtained by the formal decoder 102. On the basis of this, a reference image obtained from the VOP memory 112 and the image of the texture decoder 106 are combined to produce a single image.

Then, the edge noise remover 114 removes the edge noise and then removes the grid noise to use as the output of the decoder. Since the image input to the VOP memory 112 should be the same as that of the encoder, the image is stored without performing edge noise compensation.

As such, when the edge noise is removed by the post-processing technique, it is possible to simply apply the decoder to the decoder without changing the encoder.

A detailed process of the edge noise removing block generating the post-processed image signal by removing the edge noise from the edge noise removing unit 114 is shown in FIG. 3.

Edge noise compensation can be divided into detection process and compensation process. The detection process uses the following characteristics of blocks with edge noise.

(a) The difference between the pixel values and the blocks vertically and horizontally adjacent to the block boundary crossing point is large.

(b) It is a very smooth area around the block boundary intersection.

In FIG. 3, the condition of the difference of the average value> TH1 is for detecting the characteristic of (a), and the condition of complexity <TH2 is for detecting the characteristic of (b).

In order to analyze the noises existing over the adjacent blocks, one-dimensional pixel vectors of selected pixels of adjacent blocks are used as shown in FIG. 2, and are present across adjacent blocks in an image encoded in block units by an encoder. A method for removing block boundary noise components using post-processing is provided.

In order to quantify the degree of blocking phenomenon, the block discontinuity size is defined as a difference between an average value of four pixels of a predetermined block and an average value of neighboring blocks as shown in FIG.

That is, by defining the amount of block discontinuous threshold (Threshold: TH1, TH2), if the difference between the average value of the pixels representing each block is smaller than the threshold (TH1) or the complexity is greater than the threshold (TH2), compression It is more reasonable to think that the edges of the image span the boundary rather than the discontinuities due to the block, so that the block protects the edge part of the image by omitting post-processing by the proposed deblocking method. That is, if the difference in the average value of each block is smaller than the boundary value TH1 or the complexity is larger than the boundary value TH2, it is determined by the edge, otherwise, it is determined by the block discontinuity.

Therefore, if the block boundary is due to the edge, the decoded video is output as it is, and if the block is discontinuous, the post-processed video signal is output.

As described above, when a block that satisfies both (a) and (b) characteristics of edge noise is detected, edge noise compensation is performed on the block.

Since the boundary of the image is moved to the boundary of the block due to the edge noise, the boundary of the image should be smoothly connected using pixel values of adjacent blocks.

As shown in FIG. 1, the edge noise mainly occurs on the image boundary in the diagonal direction, and this characteristic should be reflected in the compensation process.

In FIG. 3, since the input of the edge noise removing process is one image, the block boundary crossing points to be detected are sequentially changed. (S10) Then, the average value and the complexity of the corresponding block are calculated. (S20) Average of each block It is shown in [Equation 1] about how to obtain.

[Equation 1]

Avg_A = (A ₁ + A ₂ + A ₃ + A ₄ + 2) / 4

Avg_B = (B ₁ + B ₂ + B ₃ + B ₄ + 2) / 4

Avg_C = (C ₁ + C ₂ + C ₃ + C ₄ + 2) / 4

Avg_D = (D ₁ + D ₂ + D ₃ + D ₄ + 2) / 4

_A in Avg_A represents an A block, and _B, _C, _D also mean each block. Since a value representing each block is set as an average value of several pixels, malfunctions can be prevented in the high frequency region.

Equation 2 is used as an example that can be used to calculate the complexity of each block.

[Equation 2]

Diff_A = A _1- A ₂ | + | A ₁ -A ₃ | + | A ₁ -A ₄ |

Diff_B = B ₁ -B ₂ | + | B ₁ -B ₃ | + | B ₁ -B ₄ |

Diff_C = C ₁ -C ₂ | + | C ₁ -C ₃ | + | C ₁ -C ₄ |

Diff_D = D ₁ -D ₂ | + | D ₁ -D ₃ | + | D ₁ -D ₄ |

In Equation 2, _A, _B, _C, and _D represent each block as in the case of [Equation 1].

An embodiment of the detection process will be described based on the detection process of the A block.

The difference between the adjacent B and C blocks and the average value is calculated and it is determined that the condition (a) is satisfied when the difference values exceed the boundary value TH1. This is expressed as Equation 3 below.

[Equation 3]

| Avg_A-Avg_B | > TH1 and | Avg_A-Avg_C | > TH1

In Equation 3, _A, _B, and _C represent each block as in the case of [Equation 1].

TH1, which is a boundary value used herein, is preferably determined according to QP, which is a quantization parameter, and one embodiment is shown in [Equation 4].

[Equation 4]

TH1 = 2 * QP

TH2 = 2 * QP + 20

The complexity threshold is also preferably determined according to the quantization parameter QP, and as an example, TH2 of Equation 4 may be used. If the complexity for each block is less than TH2, the condition of (b) is satisfied.

The above process is shown in [Equation 5] as an example for the A block. (S40)

[Equation 5]

Diff_A <TH2

The compensation process will be described when the A block is detected as a block in which edge noise is satisfied by satisfying both the conditions (a) and (b) (S50).

Pixels with corner noise make the image boundary appear discontinuously, so it is reasonable to compensate for the image orientation. However, to reduce the complexity of the compensation process, the compensation is performed diagonally, which is the most common form of edge noise. Thus, in the case of the A block, the A ₁ pixel should have a value similar to the B _1, B _3, C _{1 and} C ₂ pixels of the B and C blocks. The A ₂ pixels have values similar to those of the C block in which the block boundaries are adjacent, and the A ₃ pixels have similar values to those of the B block.

An example of the above-described compensation process is shown in [Equation 6].

[Equation 6]

A ₁ '= (2 * A ₁ + 2 * B ₁ + 2 * C ₁ + B ₃ + C ₂ + 4) / 8

A ₂ '= (2 * A ₂ + C ₂ + C ₅ + 2) / 4

A ₃ '= (2 * A ₃ + B ₃ + B ₆ + 2) / 4

The above compensation equation can effectively compensate for the noise, including consideration of edge noise caused by the diagonal image boundary located at the intersection of the block boundary.

The equation for the compensation of the B, C, and D blocks may be applied symmetrically to the block boundary.

5 and 6 are block diagrams of an encoding apparatus and a decoding apparatus according to an embodiment of the cyclic filter technique of the present invention.

5 is a block diagram of an encoding apparatus according to an embodiment of the cyclic filter technique according to the present invention, which includes a formal encoder 200, a motion estimator 202, a motion compensator 204, a texture encoder 206, and a VOP memory. 208, the edge noise canceller 210, and the multiplexer 212 are configured.

The corner noise canceling block of FIG. 5 has the same function as the corner noise removing block described in the block diagram of FIG. .

In FIG. 5, the input video encodes shape features of objects in the input video through the formal encoder 200, and the motion estimator 202 searches for a reference image most similar to the current coding block.

The motion compensator 204 obtains the image found by the motion estimator 202 from the reconstructed previous VOP memory 208. Thereafter, the difference between the two images is obtained and texture-coded by the texture encoder 206 to form a single encoded bit string through the multiplexer 212.

When encoding at a low bit rate, the texture encoder 206 removes a lot of high-frequency components during the above process, which causes edge noise. Then, when the edge noise is input to the VOP memory 208, the edge noise is propagated to the subsequent image to obtain an unnatural image. Therefore, if the reference image is made naturally by performing edge noise removal after texture coding, it is possible to prevent the edge noise from propagating.

In this case, unlike the post-processing technique of FIG. 4, the decoding apparatus of the cyclic filter technique of FIG. 6 may input an image from which edge noise is removed to the VOP memory 312 so that it may be decoded in the same manner as the encoder.

6 is a block diagram of a decoding apparatus according to an embodiment of the cyclic filter technique of the present invention, including a demultiplexer 300, a formal decoder 302, a motion decoder 304, a texture decoder 306, and a motion. The compensator 308, the VOP reconstructor 310, the VOP memory 312, and the edge noise canceller 314 are configured.

The encoded bit strings generated by the multiplexer 212 of the encoding apparatus are input to the demultiplexer 300 and separated into data to be input to the formal decoder 302, the motion decoder 304, and the texture decoder 306. do. After that, the shape decoder 302 decodes the shape feature of the input image, the motion decoder 304 decodes the motion vector value estimated by the encoder, and the texture decoder 306 refers to the motion vector. Decode the difference image between the image and the encoded original image.

The motion compensator 308 uses the decoded image input from the motion decoder 304 and the image from which edge noise is input from the VOP memory 312 to remove the characteristic of the image obtained from the formal decoder 302. Compensate to generate an image signal, and the VOP reconstruction unit 310 takes the position of the reference image obtained by the motion decoder 304 in the VOP memory 312 based on the characteristics of the image obtained by the stereotype decoder 302. The image is combined with the image of the texture decoder 306 to produce a single image.

Thereafter, the edge noise remover 314 removes the edge noise and then removes the grid noise to output a decoded image to be used as an output of the decoder.

Thus, when the edge noise is removed by the cyclic filter technique, the edge noise removing block described in the post-processing technique of FIG. 4 has the same function, but the edge noise is removed from the decoded image used as the reference image during encoding. Can be further increased.

As described above, according to the edge noise detection and compensation method according to the present invention, the edge noise of the inter-block boundary noise generated in the encoding process of the image coded in block units is removed, and at the point where the block boundary crosses By compensating for a pixel having a great difference from the surrounding pixel values to be similar to the surrounding pixel values, the image having the best image quality is reproduced without discontinuity between the internal pixels of the block.

In addition, there is an effect of reducing the unnatural noise of the transmitted image to facilitate the transmission of the image of a low bit rate.

What has been described above is just one embodiment for implementing the edge noise detection and compensation method according to the present invention, the present invention is not limited to the above-described embodiment, it is within the technical spirit of the present invention Of course, various modifications are possible by those who have knowledge.

1 is a diagram illustrating an example in which corner noise occurs;

2 is a diagram showing pixel positions at points where block boundaries intersect;

3 is a flowchart illustrating a detailed process of edge noise detection and compensation according to the present invention;

4 is a block diagram of a decoding apparatus according to an embodiment of a post-processing technique of the present invention;

5 is a block diagram of an encoding apparatus according to an embodiment of a cyclic filter technique of the present invention;

6 is a block diagram of a decoding apparatus according to an embodiment of the cyclic filter technique of the present invention.

 * Description of the symbols for the main parts of the drawings *

100,300: demultiplexer 102,302: orthopedic decoder

104,304: Motion Decoder 106,306: Texture Decoder

108,204,308: motion compensation unit 110,310: VOP reconstruction unit

112,208,312: VOP memory 114,210,314: Corner noise canceller

200: standard encoder 202: motion estimation unit

206 texture encoder 212 multiplexer

Claims (5)

In the method for removing the corner noise of the image generated in the encoding process of the image encoded in block units, Calculating an average value and a complexity of a block to be detected around a block boundary intersection point in a decoded image and an adjacent block vertically and horizontally adjacent to the block; And Comparing the calculated inter-block difference and complexity of the average value with a boundary value obtained using a quantization parameter to detect a block in which edge noise has occurred. The corner noise detection step, (a) comparing the difference between the average value and the vertically and horizontally adjacent blocks around the block boundary intersection point; (b) detecting whether the area is smooth around the block boundary intersection point, If a block that satisfies all of the steps (a) and (b) is detected, the edge noise detection method is defined as a block in which edge noise is generated. The method of claim 1, In the step (a), after calculating the average value of several pixels representing each block, the difference between the average value and the adjacent block is calculated to detect a case where the difference value exceeds the block discontinuous boundary value (TH1), The edge value TH1 is a corner noise detection method, characterized in that calculated by 2 * QP (quantization parameter). The method according to claim 1 or 2, In the step (b), the case where the complexity obtained through the difference of pixel values in the detection object block is smaller than the complexity threshold value TH2 is detected, and the threshold value TH2 is 2 * QP (quantization parameter) + 20. Edge noise detection method characterized in that the calculated. In the method for removing the corner noise of the image generated in the encoding process of the image encoded in block units, (c) The edge noise compensation method is characterized in that the image boundary in the diagonal direction is located at the intersection of the block boundary, so that the edge of the image is smoothly connected using the pixel value of the adjacent block. The method of claim 4, wherein In the step (c), since the corner noise occurs in the diagonal 45 degrees in the case of the A block, (c1) compensating for the A ₁ pixel by the following equation such that the block boundary has a value similar to the B _1, B _3, C _{1 and} C ₂ pixels of the adjacent B and C blocks; A ₁ '= (2 * A ₁ + 2 * B ₁ + 2 * C ₁ + B ₃ + C ₂ + 4) / 8 (c2) compensating for the A ₂ pixel by the following equation such that the block boundary has a value similar to the C ₂ and C ₅ pixels of the adjacent C block; And A ₂ '= (2 * A ₂ + C ₂ + C ₅ + 2) / 4 (c3) compensating for the A ₃ pixel by the following equation such that the block boundary has a value similar to the B ₃ and B ₆ pixels of the adjacent B block; A ₃ ′ = (2 * A ₃ + B ₃ + B ₆ + 2) / 4.