DE19747402A1 - Block based digital image processing method - Google Patents

Abstract

The method involves processing at least one image block of a digital image which includes image points grouped in image blocks, in which the image block includes a spectral coefficient which describes a common part of a coding information of the image points. The common part of the image block, and associated common parts of neighbouring image blocks are determined, and the neighbouring image blocks are divided into two groups, so that the common parts of a first group and the second group are sufficiently similar. The image block is divided into a first and a second partial image block. The first partial image block is associated with a common part which is approximately equal to the common parts of the neighbouring image blocks of the first group. The second partial image block is associated with a common part which is approximately equal to the common parts of the neighbouring image blocks of the second group.

Description

The invention relates to the processing of at least one image blocks of a digital image.

With a block-based image coding method, block artifacts especially at low transmission rates decoded pictures. Under block artifacts are discontinuous verities of coding information at the block edges stand. This is due in particular to discontinuity on the block edges in the reconstructed images, the independent, block-by-block coding of coding information from a digital image.

The coding information is further the pixels brightness information associated with an image and / or Understand color information.

As part of the image enhancement of a decoded image it is known to subject block edges to low pass filtering hen, which reduces discontinuities at the block edges become [1]. By filtering, however, in the picture existing structures, such as edges, partially broken disturbs. Thus, the reconstructed, improved image is in accordance with the process from [1] out of focus at the block edges, in particular the edges contained in the picture appear "blurred".

Various block-based image coding methods are knows, e.g. B. MPEG 1, MPEG 2 [2], H.261 [3].

The invention is therefore based on the problem of a digital one Image, which has block artifacts, for processing undergo, by processing structures in the  edited image are better preserved than with the known procedures.

The problem is solved by the method according to claim 1 and solved by the device according to claim 11.

In the method, each picture block, in which picture elements of the decoded digital picture are grouped, has a spectral coefficient which describes a constant component of coding information of the picture elements. The following method steps are carried out for at least one image block of the image:

• - The DC component of the image block is determined;
• - for picture blocks adjacent to the picture block, which are in the wide ren as neighboring picture blocks, these are assigned direct components determined;
• - The neighboring picture blocks are grouped into two groups, see above that the direct components of the neighboring picture blocks of a first Group and the equal parts of the neighboring picture blocks the second group are sufficiently similar to each other;
• - The picture block is in a first sub-picture block and in egg split a second field block;
• - The first field block is the first partial equal component assigned the DC component, which is approximately equal to one DC component of the neighboring picture blocks of the first group;
• - The second field block is the second partial DC component assigned the DC component which is approximately the same Equal share of the neighboring picture blocks of the second group.

The device for image improvement has a processor unit, which is set up in such a way that each image block, in which the pixels of the image are grouped, has a spectral coefficient which describes a constant component of coding information of the pixels. The following procedural steps are carried out for at least one image block:

• - The DC component of the image is determined;  
• - for picture blocks adjacent to the picture block which are in the wide ren as neighboring picture blocks, these are assigned direct components determined;
• - The neighboring picture blocks are grouped into two groups, see above that the direct components of the neighboring picture blocks of a first Group and the equal parts of the neighboring picture blocks the second group are sufficiently similar to each other;
• - The picture block is in a first sub-picture block and in egg split a second field block;
• - The first field block is the first DC component of Assigned equal share, which is approximately equal to an equal proportion of the neighboring picture blocks of the first group;
• - The second field block is the second partial DC component assigned the DC component, which is approximately equal to one Equal share of the neighboring picture blocks of the second group.

The invention improves the edge information tion of the reconstructed, d. H. decoded picture. The subjective quality impression of the picture for considerably increases a viewer, especially since the Viewer no longer an impression of a ver has a blurred image, as is the case with known methods of Case is.

Further developments of the invention result from the dependent Claims.

The method is significantly simplified by further training and can therefore be implemented with considerably less computing time if the following steps are provided:

• - Using the DC components of the neighboring picture blocks becomes a gradient with an indication of the direction of the gradient determined for the image block;
• - a first DC component of the neighboring picture blocks of the first Group is selected;
• - A second DC component of the neighboring picture blocks of the second Group is selected;  
• - The first partial DC component is the first DC component;
• - The second partial DC component is the second DC component.

It is also to further improve the overall impression of the decoded picture in a further embodiment partial that the size of the first sub-picture block and of the second field block results from the ratio of the first DC component and the second DC component at the.

Another image enhancement is in a further issue achieved in which a new image block fb the image block gb and the sub-image block nb according to the following font gives:

being with

• - α a predeterminable first weight factor,
• - β a predeterminable second weight factor,

referred to as.

It is advantageous if the first weight factor α and the second weight factor β are formed depending on at least one of the following criteria:

• - a difference between the first DC component and the second DC component;
• a ratio of the number of neighboring picture blocks in the first Group and the number of neighboring picture blocks in the second group pe;
• - a difference between the DC component and the first DC share.

An embodiment of the invention is shown in the figures and will be explained in more detail below.  

Show it

FIG. 1a to 1c sketches the course of the DC components prior to (Fig. 1a) and after (Fig. 1b) of the Bildver improvement for an image block and the resulting image block (Fig. 1c);

Figs. 2a and 2b an image block and eight neighboring image blocks of the image block, wherein it is shown that the adjacent image blocks are divided into two groups (Fig. 2a) and a gradient of the image block with indication of direction (Fig. 2b);

Fig. 3 is a diagram of a video encoder which operates in accordance with a block-based image encoding method;

Fig. 4 is a sketch of an image decoding unit which operates according to a block-based image encoding method;

Fig. 5 is a sketch of an image with image blocks.

FIG. 3 shows a camera 301 which takes pictures 302 of a scene and feeds them to an image coding unit 303 .

The images 302 represent a sequence of images which follow one another in time and which are successively fed to the image coding unit 303 .

Each image has pixels 501 (cf. FIG. 5), each of which is assigned coding information (brightness information or color information). The pixels are grouped into image blocks 502 of size 8 × 8 pixels 501 .

A distinction is made between block-based image coding inter-picture coding and intra-picture coding.

In the inter-picture coding, as in the rest of it purifies, only difference image information between two di directly encoded and transmitted successive images. At  the intra-picture coding becomes the entire picture information tion of an image coded and transmitted.

In inter-image coding, an image 302 to be encoded is fed to a subtraction unit 304 . In the subtraction unit 304 , a difference is formed between the coding information of the pixels of the image to be coded and the coding information of pixels of a directly preceding image that has been reconstructed in the image coding unit 303 . A difference image 305 is subjected to a discrete cosine transformation (DCT transformation) block by block in a transformation unit 306. Resulting, transformed image blocks 307 are supplied to a quantization unit 308 .

Each transformed picture block has spectral coefficients which each describe the spatial frequencies of the coding information of the respective picture blocks 502 . The spectral coefficients have a constant component and coefficients that describe components of higher spatial frequencies.

In the quantization unit 308 , the transformed image blocks 307 are quantized to form quantized image blocks 309 . Entropy coding is applied 310 to the transformed, quantized image blocks 309 , with run-length coding and variable-length coding (RLC / VLC) being provided.

The resulting encoded image 311 is transmitted from the image coding unit 303 to an image decoding unit 401 shown in FIG. 4.

The transmission takes place via any communication network (fixed network or mobile network). The quantized image blocks 309 are subjected to an inverse quantization 312 in the image coding unit 303 . The inverse quantized image blocks 313 are subjected to an inverse discrete cosine transformation (IDCT) 314 . The resulting reconstructed image blocks 315 are fed to an addition unit 316 . The information of an image block of a temporally previous image is supplied in the addition unit 316 . The image block is an image block in the previous image, which is added to the respective reconstructed image block 315 as a result of a motion estimation 317 as part of a motion compensation 318 . The resulting re-constructed overall image block is stored in an image memory 319 .

In the image decoding unit 401 , a corresponding method is carried out, as described above, for the reconstruction of the image.

Entropy decoding 402 (variable-length decoding, run-length coding, VLD / RLD) is carried out for the coded image. The entropy-decoded, quantized image blocks 403 are subjected to an inverse quantization 404 and the resulting inverse quantized image blocks 405 are subjected to an inverse discrete cosine transformation (IDCT) 406 . These resulting reconstructed image blocks 407 are fed to an adder unit 408 , as is an image block of a previous image which is also fed to the adder unit 408 as part of the motion compensation 409 .

This results in carrying out the method for all image blocks of an image, in each case a decoded image 411 stored in an image memory 410 , which has the block artifacts described above. The decoded image is subjected to image enhancement in an image enhancement means 412 . An improved image 413 shown to the viewer of the image is the result of the method which is explained in more detail below and is applied to at least some of the image blocks of the decoded image 411 .

Fig. 2a shows an image block 201 having a DC component DC _0.

It is selected for each image block 200 for which the image improvement is to be carried out, the image block 200 is selected and the method described below is carried out for the respective image block 200 .

The image block 200 has 8 neighboring image blocks 201 , 202 , 203 , 204 , 205 , 206 , 207 , 208 , each with a direct component assigned to the image block DC ₁ , DC ₂ , DC ₃ , DC ₄ , DC ₅ , DC ₆ , DC ₇ and DC ₈ on.

The neighboring picture blocks 201 , 202,. . ., 208 are grouped into two groups, a first group A and a second group B. Neighboring picture blocks are grouped together into a group if the DC component of the neighboring picture blocks is sufficiently similar.

It makes sense to use the following criterion for grouping move.

The first group A, the neighboring image blocks 201, 202, 203, 204, 205 assigned, the DC component DC ₁ to DC ₅ is exceed ing the DC component DC of the image block ₀ 200th

The second group B is supplied with the image blocks, 206 , 207 , 208 , whose direct components DC ₆ , DC ₇ , DC _{8 are} smaller than the direct component DC _{0 of} the image block 200 .

An edge possibly contained in the image block 200 and the neighboring image blocks is determined by determining a gradient vector (amount and direction of the gradient) on the basis of the equal components of the image block 200 and the neighboring image blocks in accordance with the following regulation:

being with

• - DC _k , k = 1,. . . , 8, the direct components of the eight directly adjacent neighboring picture blocks of the picture block in line-by-row numbering,
• - i , j two mutually orthogonal directions in the image.

A direction of a dividing line between neighboring picture blocks of the first group A and neighboring image blocks of the second Group B depends on the direction of the gradient vector. The direction can be vertical, horizontal, or 45 ° each be inclined to the left or right around the horizontal.

Fig. 2b shows an assumed gradient vector degree for the image block 200 and a coordinate system, spanned by two mutually orthogonal coordinate axes i and j , which are aligned according to the orientation of the image block and the bar image blocks.

The direction of the gradient vector grad is used as the basis for the further procedure. The direction of the gradient vector grad serves as the basis for a diagram shown in FIG. 1a.

The local course is plotted along the course of the direction of the gradient vector degrees on the abscissa of the diagram shown in FIG. 1a.

The direct component of the image block or neighboring image block, which is located at the corresponding local position along the gradient vector grad, is plotted on the ordinate of the diagram.

So is located in a first area 101 along the Gra serves vector degree a neighbor image block, which has a sliding part Chan DC _B. The image block 200 with the direct component DC ₀ is located at a second position 102 . S denotes an area shown in the diagram, which results from the product of the direct component DC _{0 of} the image block 200 and the size of the second interval 102 . In a third local interval is 103 degrees along the gradient vector, a neighbor image block of the first group A, which comprises a DC component DC _A.

The direct components DC _A , DC _{B of} the neighboring picture blocks remain unchanged in the further method steps. Only the direct component DC _{0 of} the image block 200 is changed in accordance with the following regulation.

The image block 200 is divided into two partial image blocks in such a way that, on the one hand, the width of the image block along the gradient vector degree 102 is retained (see FIG. 2b) and, on the other hand, the area S is also retained. The sub-image blocks are in their width along the gradient of the art in the first sub-image block, the surface level along the gradient is referred to as S1, and is divided to two th field block whose surface is denoted by S2 in the same proportions, according to the behaves nis the DC component DC _A of neighboring image blocks of the first group A and _B of the DC component DC of the neighboring image blocks of the second group B.

This means that a width of the first field block along the gradient vector a and a width b of the second field block along the gradient vector grad are formed in accordance with the following regulations:

where c denotes the width of the second interval 102 along the gradient vector grad .

In the first sub-image block with the width a along the Gra serves vector degree the DC component DC _B is assigned to the neighboring image blocks of the second group B. The DC component of the second image block, the DC component DC _A is associated with the neighboring image blocks of the second group A.

A sketch of an image block resulting therefrom, which is referred to below as an improved image block fb, is shown in FIG. 1c.

It has proven to be advantageous not always to use the method in a balanced manner for the entire block, but to carry out a further weighting of the resulting gradient block with the image block 200 to the final image block fb in accordance with the following regulation:

being with

• - α a predeterminable first weight factor,
• - β a predeterminable second weight factor,

referred to as.

The first weight factor α and the second weight factor β are formed taking into account the following criteria and according to the following regulation:

being with

• - a first summand con a difference between which he most DC component and the second DC component,
• - a second summand num a ratio of the number after bar image blocks in the first group and the number of neighboring images lock in the second group,
• - a third summand pos a difference between the DC component and the first DC component,

referred to as.

A first criterion is taken into account in a first summand. The first criterion is the contrast between the DC components DC _A of neighboring image blocks of the first Grup pe A and the neighboring image blocks of the second group B, that is, ei ne difference between the first DC component DC _A and the second DC component DC _B.

The first summand con is formed according to the following rule:

where DC _{max denotes} the maximum value of all DC components. With a value DC _max = 255, the following values result for:

One in a second summand nom of the above regulation Be The second criterion to be taken into account is a number of neighboring picture blocks belonging to the first group A or the second group B is assigned. The more neighboring picture blocks are assigned to the second group B, the larger the second weight factor β selected.

The second summand nom is formed according to the following rule:

• - The second summand nom is assigned the value 1 if a group of neighboring picture blocks 3, 4 or 5 neighboring picture blocks are assigned.
• - The second summand nom is assigned the value 0 if a group of neighboring picture blocks 1, 2, 6 or 7 neighbors image blocks are assigned.

A criterion which is taken into account in a third summand pos is the difference between the direct component DC _{0 of} the image block 200 with the direct component DC _{A of} the neighboring image block of the first group A or of the direct component DC _{B of} the neighboring image blocks of the second group B.

The third summand pos is formed as follows font:

It has proven advantageous, the first Gewichtsfak tor α assign a more lower value, the closer the DC component DC ₀ of the image block 200 to the DC component DC _B of the neighboring image blocks of the second group B is compared to the DC component DC _A of neighboring image blocks of the first group A .

The procedure described above is for at least one Part of the picture blocks of the picture performed. After done Image enhancement, the image is presented to a viewer.

In an alternative embodiment, the following regulations for the formation of the first addend con and the second addend nom have proven to be advantageous, with the names introduced above being retained in the following:

• - first summand con:
  
• - second summand nom:
  • - The second summand nom is assigned the value 1 if 4 neighboring picture blocks are assigned to a group of neighboring picture blocks.
  • - The second summand nom is assigned the value 0.6 if a group of neighboring picture blocks 3 or 5 are assigned to bar picture blocks.
  • - The second summand nom is assigned the value 0.2 if a group of neighboring image blocks 2 or 6 after bar image blocks are assigned.
  • - The second summand nom is assigned the value 0 if a group of neighboring picture blocks 1 or 7 are assigned to bar picture blocks.

The following publications were cited in the context of this document: [1] S. Minami and A. Zakhor, An Optimization Approach for Removing Block Effects in Transform Coding, IEEE Transactions on Circuit and Systems Video Technology, Vol. 2, pp. 74-82, USA, April 1995
[2] D. Le Gall, MPEG: A Video Compression Standard for Multimedia Applications, Communications of the ACM, vol. 34, no. 4, pp. 47-58, April 1991
[3] Ming Liou, Overview of the px64 kbit / s Video Coding Standard, Communications of the ACM, Vol. 34, No. 4, pp. 60-63, April 1991

Claims (20)

1. Method for processing at least one image block of a digital image which has pixels grouped into image blocks,

• in which the picture block has a spectral coefficient which describes a constant component of coding information of the picture points,
• - in which the direct component of the image block is determined,
• - in which for identical picture blocks adjacent to the picture block, according to bar picture blocks, these are determined,
• in which the neighboring picture blocks are grouped into two groups, so that the direct components of the neighboring picture blocks of a first group and the direct components of the neighboring picture blocks of a second group are each sufficiently similar to one another,
• in which the image block is divided into a first field block and a second field block,
• in which the first partial picture block is assigned the direct part, which is approximately equal to a direct part of the neighboring picture blocks of the first group,
• - In which the second part picture block is assigned as the second part equal part, the direct part which is approximately equal to a direct part of the neighboring picture blocks of the second group.

2. The method according to claim 1, where several image blocks of the digital image are processed become. 3. The method according to claim 1 or 2, in which the digital image is a decoded digital image. 4. The method according to any one of claims 1 to 3,

• in which a gradient is determined using the direct components of the neighboring image blocks with a directional indication of the gradient for the image block,
• a first DC component of the neighboring picture blocks of the first group is selected,
• a second DC component of the neighboring picture blocks of the second group is selected,
• - in which the first partial DC component is the first DC component,
• - in which the second partial DC component is the second DC component.

5. The method according to claim 4, where the size of the first sub-picture block and the size of the second sub-picture block results from the ratio the first DC component and the second DC component. 6. The method according to any one of claims 1 to 5, wherein the gradient grad is formed according to the following rule:
being with

• - DC _k , k = 1,. . ., 8 the equal components of the eight directly adjacent picture blocks in line-by-row numbering,
• i , j two mutually orthogonal directions in the image,

be designated. 7. The method according to any one of claims 1 to 6, in which a new image block fb results from the image block gb and the partial image blocks nb according to the following rule:
being with

• - α a predeterminable first weight factor,
• - β a predeterminable second weight factor,
• referred to as.

8. The method of claim 7, wherein the first weight factor α and the second weight factor β are formed depending on at least one of the following criteria:

• a difference between the first DC component and the second DC component,
• a ratio of the number of neighboring picture blocks in the first group and the number of neighboring picture blocks in the second group,
• - a difference between the DC component and the first DC component.

9. The method of claim 7 or 8, wherein the first weight factor α and the second weight factor β are formed according to the following rules:
being with

• a first summand con is a difference between the first DC component and the second DC component,
• a second summand num a ratio of the number of bar image blocks in the first group and the number of neighboring image blocks in the second group,
• a third summand pos is a difference between the DC component and the first DC component,

referred to as. 10. The method according to any one of claims 1 to 9, where the coding information is a brightness value and / or is a color value.   11. Device for processing at least one image block of a digital image which has pixels grouped into image blocks,
with a processor unit which is set up in such a way that

• the picture block has a spectral coefficient which describes a constant component of coding information of the picture elements,
• - the DC component of the image block is determined,
• for direct picture blocks adjacent to the picture block, neighboring picture blocks, associated direct components are determined,
• the neighboring picture blocks are grouped into two groups, so that the direct components of the neighboring picture blocks of a first group and the direct components of the neighboring picture blocks of a second group are each sufficiently similar to one another,
• the image block is divided into a first partial image block and a second partial image block,
• the first partial picture block is assigned the direct part, which is approximately equal to a direct part of the neighboring picture blocks of the first group, as the first partial direct component,
• - The second partial picture block is assigned as the second partial direct component the direct component, which is approximately equal to a direct component of the neighboring picture blocks of the second group.

12. The device according to claim 11, in which the processor unit is set up in such a way that multiple image blocks of the digital image can be edited. 13. The apparatus of claim 11 or 12, in which the processor unit is set up such that the digital image is a decoded digital image. 14. The device according to one of claims 11 to 13, wherein the processor unit is set up such that

• a gradient with an indication of the direction of the gradient for the image block is determined using the direct components of the neighboring image blocks,
• a first DC component of the neighboring picture blocks of the first group is selected,
• a second DC component of the neighboring image blocks of the second group is selected,
• - the first partial DC component is the first DC component,
• - The second partial DC component is the second DC component.

15. The apparatus according to claim 14, in which the processor unit is set up such that the The size of the first sub-picture block and the size of the second Sub-picture blocks each result from the ratio of he most DC component and the second DC component. 16. Device according to one of claims 11 to 15, in which the processor unit is set up in such a way that the gradient grad is formed in accordance with the following regulation:
being with

• - DC _k , k = 1,. . ., 8 the equal components of the eight directly adjacent picture blocks in line-by-row numbering,
• i, j two mutually orthogonal directions in the image,

referred to as. 17. Device according to one of claims 11 to 16, in which the processor unit is set up in such a way that a new image block fb results from the image block gb and the sub-picture blocks nb according to the following rule:
being with

• - α a predeterminable first weight factor,
• - β a predeterminable second weight factor,

referred to as. 18. The apparatus of claim 17, wherein the processor unit is set up such that the first weight factor α and the second weight factor β are dependent on at least one of the following criteria:

• a difference between the first DC component and the second DC component,
• a ratio of the number of neighboring picture blocks in the first group and the number of neighboring picture blocks in the second group,
• - a difference between the DC component and the first DC component.

19. The apparatus of claim 17 or 18, wherein the processor unit is set up such that the first weight factor α and the second weight factor β are formed according to the following rules:
being with

• a first summand con is a difference between the first DC component and the second DC component,
• a second summand num a ratio of the number of bar image blocks in the first group and the number of neighboring image blocks in the second group,
• a third summand pos is a difference between the DC component and the first DC component,

referred to as. 20. Device according to one of claims 11 to 19,  in which the processor unit is set up such that the Coding information a brightness value and / or a color is worth.