JP2962815B2 - Image processing method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a method for decoding a halftone image at a high compression rate using an image coding method of orthogonally transforming and coding in block units. The present invention relates to an image processing method for improving image quality deterioration due to block distortion caused by the above-described method, and is particularly suitable for low bit rate image transmission or image recording requiring a high compression ratio.

[Conventional technology]

In recent years, a plurality of M × N
An encoding method that performs orthogonal transformation on a block-by-pixel basis has come to be used. This is because orthogonal transform coding can achieve a higher compression rate than prediction coding that has been used in the past, and the processing of orthogonal transform itself itself was too large in computational complexity to achieve practical use. This is due to the fact that processing speeds that are practically acceptable can be obtained due to advances in LSI technology.

[Problems to be solved by the invention]

By the way, in the coding by the orthogonal transform in block units, an image with relatively little deterioration can be reproduced even if the compression ratio is considerably increased, but image quality deterioration peculiar to the orthogonal transform also occurs. As such image quality degradation peculiar to the orthogonal transformation, there is block distortion in which a boundary between processing blocks is visible in an image flat portion.
However, when a higher compression ratio is required, it is necessary to improve these image quality degradations.

In block distortion, AC components are lost due to compression in a flat part of the image where there is little change in gradation, and only DC components remain, so that gradation becomes stepwise for each processing block, and this step becomes a block boundary. It appears. Therefore, in order to remove block distortion, there has been proposed a method of estimating the gradation level values of the four corner points of each block from the decoded pixels in which block distortion has occurred, and correcting the inside of the block from the error of the estimated value ( 1989 Image Coding Symposium, 4-3
“Study on block distortion removal”). However, according to this method, correction amounts of four low-order transform coefficients of the orthogonal transform for correcting the errors at the four points are obtained from the estimated errors at the four points, and these are added to the transform coefficients at the time of the inverse transform. The correction process is slightly complicated.

The present invention improves image quality deterioration due to block distortion,
It is an object of the present invention to provide an image processing method for correcting a decoded image so that visual deterioration is less noticeable.

[Means for solving the problem]

The present invention provides a mosaic-like process that occurs in a decoded image when a halftone image is coded at a high compression rate using an image coding method that performs orthogonal transform in units of a plurality of M × N pixels and codes the block. An image processing method for removing a block boundary, the method comprising the steps of: calculating, for each corner point of each block in a decoded image, a target angle point and gradation level values of three adjacent corner points of three blocks around the target angle point; The tone level value of the corner point in the original image corresponding to the focused corner point is estimated, and an error between the estimated value and the tone level value of the square point of the decoded image is obtained. The error is linearly interpolated within the block and added to the decoded image to remove block boundaries.

(Operation)

In the present invention, in order to remove a mosaic-like processing block boundary (block distortion) generated in a decoded image by orthogonal transform coding, first, for each of the four corner points of each block, the corner points of three neighboring blocks are used. The average value of the gradation levels of the target corner point and the three neighboring corner points is taken so as to eliminate the difference therebetween, and is used as an estimated value of the gradation level of the target corner point. Subsequently, an error between the estimated value and the gradation level of the actual decoded image is obtained, and this is linearly interpolated in the block and added to the decoded image for correction. This series of processing is repeated for all blocks, and block distortion is removed by smoothly connecting block boundaries. However, when the target angle point is on the edge, the average value is obtained by excluding the triangular points adjacent to the target angle point whose gradation level difference from the target angle point is equal to or greater than a predetermined threshold value. Let it be an estimated value. By doing so, the side effect of blurring the edge portion can be avoided. Furthermore, if each side of the processing block happens to coincide with the actual edge, it is difficult to determine that this is not block distortion, so it is determined in advance at the time of encoding, and this determination result is transmitted as additional information. I do. Then, in the block distortion removal processing at the time of decoding, as for the corner points at both ends of the side determined to be the edge, the decoded value itself is used as the estimated value. Thereby, blurring of the edge portion can be avoided.

〔Example〕

FIG. 1 is a flowchart showing the processing procedure of the block distortion correction processing of the image processing method according to the present invention.

In FIG. 1, an inverse transform process (step 1) is a process of transforming image data encoded by orthogonal transform into decoded image data. If the orthogonal transform is discrete cosine transform, inverse discrete cosine transform is performed to perform decoding. I do. The decoded image data is divided into blocks for each of a plurality of N × N pixels and cut out (step 2), and a quadrangular tone level value estimation process is performed for each block (step 3).

In this quaternary point gradation level value estimation processing, in order to smoothly connect the block boundaries, each of the pentagonal points of each block is set so that there is no difference between the vertexes of the three neighboring blocks adjacent thereto. In this process, the average value of the gradation levels of the point and the corner points of the three neighboring blocks is used as the estimated value of the gradation level of the target corner point. That is,
As described above, block distortion is distortion caused by the loss of the AC component due to compression in the flat portion of the image where there is little change in gradation, leaving only the DC component, and the gradation is stepwise in each processing block. And this step appears as a block boundary. Therefore, in the process of step 3, in order to smoothly connect the block boundaries, each of the four corner points of each block is adjacent to the corner point of interest so that there is no difference between the corner points of the three neighboring blocks. The average value of the gradation levels of the corner points of the three surrounding blocks is used as the estimated value of the gradation level of the noticed corner point. FIG. 2 shows this relationship. In the figure, a solid line indicates an orthogonally transformed decoded image including block distortion, a broken line indicates a block distortion corrected image, and a black circle indicates an estimated value of a block corner point.

Subsequently, an error between the estimated value and the value in the actual decoded image is obtained for the four corner points in the block, and this is linearly interpolated in the block (step 4), and corrected by adding to the decoded image (step 5). . Then, the series of processes (steps 2 to 5) are repeated for all blocks (step 6).

However, if the block corner point happens to be on the edge, if the value of this corner point is estimated with the average value of adjacent corner points and linear interpolation is applied, the edge part will be blurred. Gives side effects. Therefore, when the target angle point is on the edge, the estimation method in step 3 is changed, and the difference in the gradation level between the target angle point and the target angle point among the three points adjacent to the target angle point is equal to or greater than a predetermined threshold value. In the case of those that are excluded, the average value is used as the estimated value. By doing so, the side effect of blurring the edge portion can be avoided. FIG. 3 shows this relationship. In the figure, a solid line portion indicates an orthogonally transformed decoded image originally having an edge at a block boundary, a broken line portion indicates an image in which an edge portion is blurred due to block distortion correction, and a black circle indicates an estimated value of a block corner point.

Next, the actual processing of steps 3 to 5 will be described in more detail with reference to FIG. In FIG. 4, the hatched block is defined as a target block, and the gradation levels of the four corner points of the target block are set as P _ij (i, j = 0.1). Let the triangular points of the block be A _ij , B _ij , and C _ij , respectively. Here, each block is a square block and the size is a plurality of N × N pixels. In this state, the estimated values Q _ij of the four corner points of the block of interest are
It is obtained by the following equation (step 3).

However, k [0], k [1], and k [2] are defined as follows. Becomes

Next, a difference R _ij (= Q _ij −P _ij ) between the estimated value Q _ij and the value P _ij of the actual decoded pixel is obtained, and the difference R _ij is linearly interpolated in the block of interest to obtain a correction amount S _pq (p, q = 0, 1, 2, ..., N-1) (step 4).

The correction amount _Spq obtained in this way is added to the decoded image to perform correction (step 5).

Therefore, if the gradation level of each pixel of the decoded image in the block is U _pq , and the gradation level of each pixel in the block after the correction by this processing is V _pq , then V _pq = U _pq + S _pq .

Although the threshold value Th1 is set, the threshold value Th1 discriminates a false edge due to block distortion from an original edge, but is not uniquely determined. However,
Since block distortion is considered to occur due to loss of the AC component where the AC component other than the DC component is originally small in the image flat part, the step caused by the block distortion is at most the quantization step size of the DC component in the decoded image. It is considered that the step amount is equivalent to about 2-3 times. In a place where a step is formed more than this, the primary AC component is so strong that it is not lost by compression and block distortion hardly occurs. Therefore, it is reasonable to think that such a large step is due to an original edge. Therefore, it is appropriate to set the threshold Th1 to a step amount corresponding to at most about two to three times the quantization step size of the DC component in the decoded image.

However, even if such adaptive processing is performed, for example, as shown in FIG. 5, if the processing block happens to coincide with the checkerboard pattern of the original image,
It is difficult to determine that this is not block distortion by local processing. Therefore, in order to avoid such side effects, when encoding image data, it is determined in advance whether or not each side of the processing block coincides with an actual edge beforehand, and this determination result is used as additional information. Add to the image encoded data of the block. Then, for the corner points at both ends of the side of the processing block determined to be the actual edge at the time of the block distortion removal processing on the decoder side, the decoded value of the corner point itself is used as the estimated value.

For example, when each pixel of an original image in a processing block before encoding on the encoder side is K _ij (0 ≦ ij ≦ N) and each pixel of a block adjacent to the left side of the processing block is L _ij , Attention is paid to the left side of the block _Kij , and it is determined whether or not this side matches the edge. Block L This determination method is adjacent to the left of the block and the left side of the target block K _ij
If the width of change in the value of each of the N pixels on the side of _ij is within a certain threshold and the sum of the absolute values of the differences between the pixels on the two sides is equal to or greater than another threshold, this side Is determined to match the edge.

Then, it is determined that the side _Koj is an edge. Then, the decoder uses the decoded value itself as an estimated value for the corner points at both ends of the side determined to be an edge in the block distortion removal processing. That is, Q _ij = P _ij . Thereby, blurring of the edge portion can be avoided.

〔The invention's effect〕

According to the present invention, it is possible to extremely effectively improve image quality deterioration due to block distortion caused by orthogonal transform coding, and to provide a decoded image in which visual deterioration is less noticeable.

[Brief description of the drawings]

FIG. 1 is a flowchart showing a processing procedure of a block distortion correction process of an image processing method according to the present invention, FIGS. 2 and 3 are graphs showing block distortion, and FIG. 4 is a diagram showing block corner points and adjacent angles. FIG. 5 is a diagram showing an example in which the boundary of the processing block coincides with the edge of the original image;

Claims (4)

(57) [Claims] When a halftone image is encoded at a high compression ratio using an image encoding system for orthogonally transforming and encoding a plurality of M × N pixels in blocks, a mosaic-like image generated in a decoded image is obtained. What is claimed is: 1. An image processing method for removing a processing block boundary, comprising: a point of interest for each corner of each block in the decoded image; and gradation level values of three corners adjacent to the three blocks surrounding the point of interest. To estimate the gradation level value of the corner point in the original image corresponding to the noted corner point, find the error between the estimated value and the gradation level value of the four corner points of the decoded image, An image processing method comprising linearly interpolating the error at a corner point within the block, adding the error to a decoded image, and removing the block boundary. 2. An estimated value of a gradation level value of the corner point,
2. The image processing method according to claim 1, wherein an average value of a total of four gradation level values of a target corner point and three adjacent corner points of three blocks around the target corner point is used. 3. An estimated value of the gradation level value of the corner point,
Among the gradation level values of a total of four corner points of the attention corner point and three adjacent corner points of three blocks around the attention corner point, a difference between the gradation level value with the attention corner point is a predetermined threshold value. 2. The image processing method according to claim 1, wherein an average value of only the corner points excluding the above is used. 4. At the time of said encoding, it is determined whether or not each of the four sides of each processing block coincides with an actual edge, and the result of the determination is decoded as additional information together with the encoded image data. Side, and when estimating the gradation level value of the corner point, the estimated value of the corner points at both ends of the side of the processing block determined to match the actual edge is the decoded value of the corner point itself For the other corner points, as the estimated value, of the target corner point and the gradation level values of three adjacent corner points of three blocks around the corner point, a total 2. The image processing method according to claim 1, wherein an average value of only the points excluding those having a difference in tone level value equal to or greater than a certain threshold is used.