JPH0267688A - Image noise removing system - Google Patents

Abstract

PURPOSE:To prevent the generation of blur by controlling the characteristics of a filter based on relation to the position or boundary of a picture element in the case of filtering the picture element positioned on the boundary of blocks. CONSTITUTION:An input image signal is divided into blocks by a block dividing part 31, coded by a coding part 32 in each block, the coded signals 37 are decoded by a decoding part 33, decoded signals are synthesized by a block synthesizing part 34, filter characteristics are controlled by a block boundary adaptive filter part 35 in accordance with the relation with the position or boundary of the image positioned on the boundary, and an obtained output image signal 38 is outputted. The level value of an objective picture element A in a filter ring is x(0,0), the level of a peripheral picture element is x(i,j) and i, j express the relative positions from A. The level y0 of the A filtered by using the A and eight peripheral picture elements is expressed by another formula by setting up weight as f(i,j). A threshold is defined as (t) and y0 or x(0,0) is outputted in accordance with ¦x(0,0)-y0¦<t or ¦x(0,0)-y0¦>=t.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention is applicable to an orthogonal transform encoding method that divides an image into blocks and encodes each block as a unit, and an image encoding method that uses vector quantization. The present invention relates to an image noise removal technique that removes generated block noise and obtains a high-quality decoded image.

[Conventional technology]

By dividing the image into blocks, orthogonal transform coding,
It has become possible to use methods with high coding efficiency, such as vector quantization, and efforts have been made to improve image coding efficiency. However, these encoding methods produce block-like noise in decoded images. Block noise is observed when unnatural steps in brightness, color, etc. occur on the boundaries of blocks, so a smoothing filter that reduces these steps is applied to the entire decoded image or to pixels located at block boundaries. It is used to reduce blocky noise.

[Problem to be solved by the invention]

However, applying a filter to the entire decoded image will delete information other than block noise, such as edges, which the image originally has, and even if you apply a filter only to the block boundaries, If there are edges or line segments on top, the image will become blurred. That is, steps in brightness and color occur on the block boundaries on the reproduced image, giving an unnatural impression to the viewer.

[Means for Solving the Problems] In order to solve these problems, in the present invention, when performing filter processing on pixels located on the boundary of a block, the characteristics of the filter are changed depending on the position of the pixel and the boundary. It is adaptively controlled based on the relationship to prevent blurring caused by the filter.

[Operation] Generally, in a flat area, steps at block boundaries are easy to see, and even minute steps are recognized as block-like noise, so block boundaries are smoothed by filtering using pixels surrounding the filter target pixel. Furthermore, for edges, line segments, etc., the direction is detected and one-dimensional filter processing is performed in that direction. However, if the filter target pixel is on a block boundary, filtering in that direction is not performed.

Furthermore, it can be determined from the received code information whether the area is likely to generate block noise and requires filtering or not, thereby preventing unnecessary blurring. That is, in the case of encoding using motion compensation, block boundary noise is likely to occur at boundaries between blocks with different motion vectors from adjacent blocks or at block boundaries of blocks encoded as motion 8JIIM. However, if the interframe difference signal of the block is large and includes high frequency components, the blurring caused by the filter may actually degrade the image quality. Also, in orthogonal transform encoding,
It can be seen that blocks in which frequencies with high coefficients are transmitted include many edges and line segments.

Therefore, by using a block boundary filter only in the portion selected based on these criteria, block noise can be removed without causing unnecessary blurring.

〔Example〕

FIG. 3 is a diagram showing the positioning of the invention of the present application, and 31
is a block dividing unit, 32 is an encoding unit, 33 is a decoding unit,
34 is a block synthesis unit, and 35 is a block boundary adaptive filter. The block boundary adaptive filter 35 removes block noise from the decoded image and outputs an output image.

Let x (IIJ) be the value of the reference value, and i and j are the relative positions from A. Let yo be the level value of A filtered using A and the surrounding eight pixels.

- For example, the value of yo sets the filtering weight to f (
i, j), it is calculated as follows.

If the level difference does not change much before and after filtering, it is a flat area of the image, so 1x
(0,0)-y. If l is L, y0 is output, and if the level difference is large, it is an edge or line segment, so l
If x(0,0)-yo |≧t, x(00) is output.

In the above example, eight pixels are used as peripheral pixels, but any number of pixels may be used as long as isotropic filtering is achieved.

FIG. 1(a) shows the implementation of the invention of the first claim of the present application.
FIG. 1(b) is a diagram illustrating an embodiment of the invention of the second claim of Kazumoto Motoyama, and is a diagram illustrating an example of filtering target pixels, with filtering target pixels A, A, and A.
Let the level value of x (0+O) be the level of the surrounding pixel A
level value x (0, 0), level value x (i,
j) is the same as in FIG. 1(a). Let's take four radial line segments centered on A as an example. For example, if the level after filtering using pixels on line segment l is y(1), then the value of y(1) is as follows when the filtering weight is F(i, j). calculated.

Among the four line segments, the level difference between before and after filtering does not change much when the line segment matches an edge or line segment of the image. Therefore, the level value after filtering using pixels on the first line segment is Y (
i), Y (i)-x (0,0) by z(i
), 4 z (i), the smallest one is Z, giving Z y (i) is Y, threshold is L, if Z<L, smoothing in edge direction using Y after filtering as output and
If Z≧t, the value of the filtering process is not used and x(0,0) is output in order to preserve the edge. Above, we have shown an example of filtering using three pixels on a line segment, but if we want to achieve directional filtering that preserves the edges in the line segment direction, how many pixels do we need? It doesn't matter if there is. Furthermore, it is clear that the number of line segments is not limited to four.

FIG. 2 is a flowchart showing an embodiment of the invention of the third claim of the present application. This invention is a combination of the above-mentioned claims 1 and 2, and first performs isotropic filtering, and if there is a risk that the edges of the image will disappear, directional filtering is performed in the next step. Perform filtering.

Thereby, -layer effective processing can be realized.

The invention of the fourth claim of the present application, in the invention of the second claim, does not use pixels on the block boundary where the filtering target pixel is located when the filtering target pixel is located at a corner or side of the block. It is characterized by this.

In other words, when target pixel A is located on a side of a block, an nm tree is created by excluding m line segments in the same direction as the side where A is located from among 0 radial line segments centered on A. Let y(i) be the output filtered using the pixels on the first line segment among the line segments, and y(iL x (0,0
) is the absolute value of z (i), and nm pieces of z (i
), the invention of the second claim or the invention of the third claim is carried out by setting y(i) at which z (i) is the minimum as the level value after filtering. If the target pixel A is located at a corner of a block, line segments in the same direction as all sides including that pixel may be excluded from filtering, or in the case of a corner, the above fourth claim may not apply at all. It is not necessary to invent the invention.

Generally, from code information, it is possible to determine whether a region is likely to generate block noise and requires filtering or not, and thereby filter characteristics can be changed efficiently. For example, in the case of motion-compensated interframe coding, noise at block boundaries is likely to occur at boundaries between blocks with different motion vectors from adjacent blocks or at block boundaries between blocks encoded as motion areas. However, if the inside of the block contains high frequency components, the noise at the block boundary will not be noticeable, and the blurring caused by the filter may actually reduce the image quality, so it is better not to filter such blocks. For example, in orthogonal transform coding, the frequency components included in a block can be determined by the frequency of the coefficient with the highest frequency among the transmitted coefficients. Therefore, by using a block boundary filter only in the portion selected based on these criteria, block noise can be removed without causing unnecessary blurring.

FIG. 4 is a flowchart showing an embodiment of the invention of claim 5 of the present application, and shows how block boundaries are handled in the case of interframe coding with motion compensation. When the motion vectors of the input block A and the adjacent block B across side a are different, or when A is encoded as a moving area and there is a highest non-zero frequency of the orthogonal transform coefficient to When it is below the threshold value (that is, the spatial frequency is high), filtering of a is performed, and when neither of these conditions apply, control is performed so that filtering of a is not performed.

FIG. 5 shows an embodiment in which the present invention is used for motion-compensated interframe coding, in which 501 is a motion-compensated interframe prediction unit, 502 is a block-based orthogonal transform coding unit, and 50
3 and 508 are orthogonal transform decoding units, 504 and 509 are motion compensation synthesis units, 505 and 510 are frame memories, and 506
507 is a code combining unit, and code separating unit 511 is a block boundary adaptive filter unit of the present invention. The stage before code combining section 506 corresponds to the encoding side, and the stage after code separating section 511 corresponds to the decoding side. This works as follows.

That is, the parts other than the block boundary adaptive filter section 511 are a general motion-compensated interframe coding device, and the motion-compensated interframe prediction section 501 encodes the input image signal 521 and the frame memory 505 before they are stored. A difference image signal 522 is generated from the decoded signal of the encoded frame. The differential image signal 522 is converted into an orthogonal transform encoded signal 523 by the orthogonal transform encoder 502 in units of blocks. The code synthesis unit 506 synthesizes the orthogonal transform encoded signal 523 and the motion vector signal 522 and outputs an output signal 525. The output signal is separated into an orthogonal transform encoded signal 526 and a motion vector signal 527 in a code separator 507 . Orthogonal transform decoding units 503 and 508, 504 and 50
The motion compensation synthesis unit 9 and the frame memories 505 and 510 each operate in the same manner with the motion vector signal and the orthogonal transform encoded signal as input. The orthogonal transform encoded signal is converted into a decoded difference signal in the orthogonal transform decoder.

Furthermore, a decoded image signal is generated by the motion compensation synthesis unit from the decoded difference signal, the motion vector signal, and the previous decoded signal stored in the frame memory. Here, the decoded image signal is accompanied by block-like noise generated by block unit encoding. Therefore, the decoded image signal, the orthogonal transform encoded signal, and the motion vector signal are input to the block boundary adaptive filter section 511.

FIG. 6 shows an embodiment in which the block boundary adaptive filter of the present invention is used as an in-loop filter for encoding. however,
Only the encoding side is shown, and the decoding side is omitted because it has the same configuration as a part of the encoding side. Components that operate in the same way as in FIG. 5 are designated by the same numbers. 600 is a block boundary adaptive filter unit corresponding to 511 in FIG. The difference from the case of FIG. 5 is that interframe encoding is performed between the input image and the filtered image one frame before.

FIG. 7 shows block boundary adaptive filters 511 and 600.
700 is a determination circuit, 701 is a filter processing section, and 702 is a switching switch. A decision circuit 700 determines whether or not to use a filter for each block edge of the decoded image for each edge of the block. The determination method is as shown in the flowchart of FIG.

[Effects of the Invention] As explained above, according to the present invention, block-like noise that occurs in a decoded image of an image encoding method that divides an input image into blocks and encodes each block as a unit can be reduced by lines near block boundaries. Minutes and edges can be removed without blurring them.

Furthermore, it is clear that a high-quality decoded image can be obtained by combining the present invention with other filters and performing filtering with different characteristics for pixels at block boundaries and pixels other than the blocks.

[Brief explanation of the drawing]

FIG. 1 is a drawing explaining the invention of claims 1 and 2. 2. A flowchart explaining the invention of claim 3. FIG. 3 is a drawing showing the position of the invention of the present application. FIG. 4 is a flowchart explaining the invention of claim 5. FIG. 5 is a diagram showing an embodiment in which the present invention is applied to interframe coding with motion compensation. FIG. 6 is a drawing showing an embodiment in which the block boundary adaptive filter of the present invention is used as an in-loop filter for encoding. FIG. 7 is a diagram showing an example of the configuration of a block boundary adaptive filter section. 521 Human-powered image signal 522 Difference image signal 523 Orthogonal transformation code signal FIG. 5

Claims (5)

[Claims] (1) In an image encoding method that divides an image into blocks and encodes each block, an image noise removal method that performs filtering processing on pixels located at the boundaries of blocks of a decoded image, the filtering being , when the target pixel for filtering is A, the level value of A is x, the level value of A after filtering using surrounding pixels is y, and the threshold is t, if |x-y|<t, y is the output. , |x−y|≧t, an image noise removal method is characterized in that x is output when |x−y|≧t. (2) In an image encoding method that divides an image into blocks and performs encoding on a block-by-block basis, an image noise removal method that performs filtering processing on pixels located at the boundaries of blocks of a decoded image, the filtering being , the filtering target pixel is A, the level value of A is x, and the level value of A after filtering using the pixel on the i-th line segment among n radial line segments centered on A is y. (i
), the absolute value of y(i)-x is Z(i), the smallest among n z(i) is Z, y(i) giving Z is Y, and the threshold is t, then Z< An image noise removal method characterized in that when t, Y is output, and when Z≧t, x is output. (3) In an image encoding method that divides an image into blocks and performs encoding on a block-by-block basis, an image noise removal method that performs filtering processing on pixels located at the boundaries of blocks of a decoded image, the filtering being , the target pixel for filtering is A, the level value of A is x, the level value of A after filtering using surrounding pixels is y_0,
Let y(i) be the level value of A after filtering using the pixel on the i-th line segment among n radial line segments centered on A, and Z be the absolute value of y(i)-x. (i), n pieces Z(
The smallest of i) is Z, y(i) that gives Z is Y,
When the two thresholds are t_1 and t_2, |x−y_0
If |<t_1, y_0 is the output, |x−y_0|≧
An image noise removal method characterized by outputting Y when t_1 and Z<t_2, and outputting x when |x−y_0|≧t_1 and Z≧t_2. (4) In an image encoding method that divides an image into blocks and encodes each block, this is an image noise removal method that performs filtering processing on pixels located at the boundaries of blocks of a decoded image. An image noise removal method characterized by not using pixels other than the target pixel among pixels on a block boundary where the target pixel of filtering is located. (5) In an image encoding method that divides an image into blocks and encodes each block, this is an image noise removal method that performs filtering processing on pixels located at the boundaries of blocks of a decoded image. An image noise removal method characterized by changing its characteristics using code information.