KR101004612B1 - Apparatus for post-filtering of DCT coded image and method using the same - Google Patents

Abstract

Disclosed are an apparatus and method for post filtering a DCT based image. The blocking detection unit sequentially selects a predetermined number of pixels in each block from the boundaries of adjacent blocks in a direction perpendicular to the boundaries of the adjacent blocks of the decoded image, and belongs to different blocks among the selected pixels and is adjacent to each other. An output for reducing the blocking effect of the blocks by a predetermined decision rule based on the input membership function selected for the second and third input variables, which are the change values of pixels belonging to the same block, respectively, the first input variable, which is the difference value of? Determine the membership function. The interpolator interpolates pixel values of pixels selected by the filter corresponding to the determined output membership function. According to the present invention, it is possible to effectively remove the general blocking phenomenon of the compressed image.

Compression, DCT, Blocking, Interpolation, Membership Functions

Description

Apparatus for post-filtering of DCT coded image and method using the same}

The present invention relates to an apparatus and method for post-filtering a DCT-based image, and more particularly, to a post-filtering apparatus and method for detecting and removing a blocking phenomenon occurring in a DCT coded and compressed image. .

Block conversion based on image and video coding is widely used in many standards such as JPEG, H.261, MPEG-2 and H.264 / AVC. These standards use the Discrete Cosine Transform (DCT), which is known for its good energy compression characteristics and various efficient hardware / software implementations. However, DCT coded images have a problem in that there is a visual defect at high compression. Blocking is a major visual flaw in DCT based on image coding. This problem is due to the inappropriate amount of DCT coefficients when low bit rate compression is the target. In this case, small DCT coefficients are encoded to meet the needs of limited bandwidth or storage capacity. Therefore, some blocks may have a uniform intensity value, and some discontinuities may appear between the blocks.

Many techniques have been proposed to reduce the blocking defects of DCT based on image coding (ie JPEG) for a given bit cost. Some techniques suggest twjd in the transformation. For example, in a paper entitled "Deformed Discrete Cosine Transform in Image Compression and Its Applications," a Warped Discrete Cosine Transform (WDCT), which changes the transform according to the characteristics of the input signal for improved performance, Presented. An important advantage of this technique is that it implements the existing DCT over the combined block proposed in the paper "Image Transform Coding Using Inserted Blocks" and the Combined Transformation proposed in the paper "New Techniques to Reduce the Blocking Effect of Transform Coding". It can be implemented using hardware or software. Other groups of techniques aim to change the quantization method. For example, the performance of a basic JPEG can be improved using a human visual system based on the quantization table design and an optimal quantization matrix design. The recently proposed technique shows that bit planes based on effective DCT coefficient coding provide better performance than JPEG2000 using wavelet transform and are significantly better than basic JPEG.

Post-filtering is an important tool in improving the visual quality of DCT-based coded images without significant changes to the encoder or decoder. These techniques focus primarily on mitigating the blocking effect itself. For example, an iterative algorithm based on projection on convex sets (POCS) has been proposed to reduce blocking effects. Other high-speed POCSs use a new smoothness constraint set and improved quantization suppression techniques to improve performance. For example, the proposed technique in the paper entitled "Method of Reducing Blocking Effect of JPEG Image by Signal Adaptive Filtering" applies directional 1-D filtering on edge region to reduce the blocking effect while preserving edge. Then, 2-D adaptive average filtering is applied to the flat region. Based on the same concept, the proposed deblocking technique in a paper titled "de-blocking filter for low bitrate MPEG-4 video" basically shows the image area as 'smooth', 'medium'. intermediate 'and' complex 'and appropriate filtering is performed according to the complexity of the region. H.264 / AVC, a recent video coding standard, also uses adaptive block rejection filters to improve the visual quality of image frames at low bit rates.

SUMMARY OF THE INVENTION The present invention has been made in an effort to provide an apparatus and method for post-filtering a DCT-based image that can effectively remove a blocking phenomenon occurring in an image in the field of image processing.

Another technical problem to be solved by the present invention is to provide a computer-readable recording medium that records a program for executing a DCT-based image post-filtering method that can effectively remove a blocking phenomenon occurring in an image in an image processing field. There is.

A preferred embodiment of the DCT-based image post filtering apparatus according to the present invention for achieving the above technical problem, each block from the boundary of the adjacent blocks in a direction perpendicular to the boundary of the adjacent blocks of the decoded image Selects a predetermined number of pixels in sequential order, and the second and third inputs that are change values of pixels belonging to the same block and the first input variable which is a difference value between pixels adjacent to each other belonging to different blocks among the selected pixels A block detection detector for determining an output membership function for reducing the blocking effect of the blocks based on a predetermined decision rule based on an input membership function selected for a variable; And an interpolation unit which interpolates pixel values of the selected pixels by a filter corresponding to the determined output membership function.

A preferred embodiment of the method of post-filtering a DCT based image according to the present invention for achieving the above technical problem is that each block from the boundary of the adjacent blocks in a direction perpendicular to the boundary of the adjacent blocks of the decoded image. Selects a predetermined number of pixels in sequential order, and the second and third inputs that are change values of pixels belonging to the same block and the first input variable which is a difference value between pixels adjacent to each other belonging to different blocks among the selected pixels Creating a variable; Determining an output membership function for reducing the blocking effect of the blocks by a predetermined decision rule based on an input membership function corresponding to the input variables; And interpolating pixel values of the selected pixels by a filter corresponding to the determined output membership function.

The apparatus for post-filtering a DCT-based image and the method thereof according to the present invention may be applied to a digital convergence platform to provide an environment for multimedia, network, digital broadcasting, and the like, and may provide an integrated function. In addition, by using the linear interpolation method, which is the simplest operation that corrects the blocking effect, the blocking effect is corrected, and the blocking effect of the image compressed at a high compression rate can be effectively eliminated with a small operation. It is also possible to improve the performance of JPEG compressed images by 1.5 dB for lower bit rates while the gain is limited for higher bit rates. Furthermore, the apparatus and method for post-filtering a DCT-based image according to the present invention can be applied to an electronic product such as a mobile phone in which a problem of blocking defects due to limited bandwidth exists.

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the apparatus and method for post-filtering DCT-based image according to the present invention.

1 is a block diagram showing the configuration of a preferred embodiment of the apparatus for post-filtering a DCT-based image according to the present invention.

Referring to FIG. 1, a post filtering apparatus of a DCT based image according to the present invention includes a block detection detector 110 and an interpolator 120.

The blocking detection unit 110 determines an amount of blocking by an input variable generated based on pixels of a block boundary of the decoded image. In this case, the input variables I ₁ , I _2, and I ₃ for the pixels of the block boundary as shown in FIG. 2 are defined as follows.

The first input parameter in Equation 1 I ₁ represents the amount of blocking with the absolute value of the difference between the degree of the pixel designated as D and E in FIG. The other two input variables I ₂ and I ₃ are the variance values of the left and right sides (or top and bottom), respectively, at the block boundary. These three input variables are used to determine the uniform intensity for one block.

Blocking defects are mainly observed when the amount of DCT coefficient is not sufficient. In this case, the image blocks have a uniform intensity value that can result in a sudden intensity change between adjacent image blocks. Based on this observation two indicators of blocking defects can be used in the blocking detector. The first indicator is the intensity difference at the block boundary (eg, the difference between pixels named D and E in FIG. 1). The second indicator is the change in intensity between the pixels in the block. This second indicator can be simply evaluated using a block gradient.

The blocking detection unit 110 is a fuzzy logic-based detector to determine the amount of blocking in consideration of the change in intensity at the block boundary and the change in pixels at both sides of the block boundary. At this time, the blocking detection unit 110 determines the presence or absence of the blocking effect by the magnitude of the difference value of the first input variable I ₁ . If the blocked detector 110 as an example is I ₁ is 0, a determination is made that there is no blocking effect, if I ₁ is small, determination is made that the blocking effect, and it is determined that the actual boundary than the I _1, the blocking effect is greater. In addition, the block detection unit 110 determines a range of blocking effects in two adjacent blocks by the _second and third input variables I ₂ and I ₃ . That is, if the change value of the pixels selected in each block is 0, it is not included in the range of the blocking effect, and it is determined that the range of the blocking effect is formed according to the magnitude of the change value. In this case, the block detection detector 110 determines a corresponding membership function based on the range value of the input variable to which the gradient of the pixels in the block belongs. In addition, the block detection unit 110 selects an appropriate filter by associating the intensity of the block (ie, the difference between the D and E pixels in FIG. 2) and the value of the quantization parameter after the type of the region is determined.

 Fuzzy logic provides an effective method for dealing with nonlinear systems and is therefore widely used in image processing applications such as edge detection, image shading reduction and image sequence stabilization. In the first stage of the fuzzy logic based system, the fuzzy phase, the real world input variables are converted to fuzzy values using the input membership function. The fuzzy inference engine then provides fuzzy output based on rule-base. De-fuzzy is the final step in this process of transforming the results obtained in the previous step into real world variables using the output membership function for each output. In order to get good performance from fuzzy logic based systems, input / output variables, membership functions, and decision rules must be carefully selected.

The output of the block detection unit 110 as described above represents a form of spatial filtering to be applied. Gaussian membership functions are used for both input and output variables. 3A and 3B show membership functions (MFs) of input and output variables, respectively. 3A and 3B, seven membership functions are used for output variables, while four membership functions are used for input variables in the range of 0-50. Four membership functions for input variables correspond to input variables of 'low', 'medium low', 'medium high', and 'high', respectively.

The blocking detector 110 uses a decision rule to determine the type of membership function corresponding to each input and output variable. These decision rules are determined experimentally, examples of decision rules are shown in the following table. The fuzzy logic detector applied to the block detection unit 110 is a system that determines output values according to respective input values and decision rules.

Rule number I ₁ I ₂ I ₃ Print One MF1 None None MF1 2 MF2 MF3 MF3 MF1 3 MF2 MF4 MF4 MF1 4 MF4 not MF1 not MF1 MF1 5 MF2 not MF1 not MF1 MF2 6 MF3 not MF1 not MF1 MF2 7 not MF1 not MF1 MF1 MF3 8 not MF1 MF1 not MF1 MF4 9 not MF1 MF3 MF3 MF5 10 not MF1 MF2 MF2 MF6 11 not MF1 MF1 MF1 MF7

Referring to Table 1, in the case of decision rule 1, since the strength difference is low, the output membership function is determined as MF1. In the case of decision rules 2 and 3, since the variance in the block is high but the strength difference is low / medium (meaning no blocking has occurred), the output membership function is also determined to be MF1. In the case of decision rule 4, since the difference in strength at the block boundary is not high and the variance in the block is not low, it is determined that the edge exists at the block boundary but does not include the blocking effect, so the output membership function is MF1. Is determined. In the case of decision rules 5 and 6, there is a constant strength difference at the block boundary, but since the variance is not low, the output membership function is determined as MF2. In this case, some filtering may be performed. In the case of decision rules 7 to 11, since the input variable I ₁ is not low (ie not MF1), there may be a block effect on the values of the input variables I ₂ and I ₃ . In the case of decision rule 7, since I ₃ is low while I ₂ is not low, it can be said that there is some blocking on the right side (or the bottom side). Decision rule 8 is similar to decision rule 7, in which case it can be seen that there is some blocking on the left side (or on the upper side). In the case of decision rules 9 to 11, the strength difference is not low, and the variance value of the block decreases. This means that the blocking effect is increased through 1 ₁ in decision rule 9, so the amount of filtering must be increased.

The interpolator 120 performs different kinds of filtering according to the characteristics of the pixels to be filtered. A simple 1-D linear interpolator is used to enable high speed implementation. In the interpolation process, the pixel value must be changed at the block boundary according to the blocking information. And different kinds of filtering are performed for each kind of output membership function. The filtering equation associated with the output of the block detection unit 110 is given by the following equation.

Where A, B, C, and D are pixel values of pixels sequentially selected in one adjacent block, and E, F, G, and H are pixel values of pixels sequentially selected in another adjacent block, and A ', B', C 'and D' are interpolated pixel values of pixels sequentially selected in one adjacent block, respectively, and E ', F', G 'and H' are selected sequentially in another adjacent block, respectively. The interpolated pixel value of the pixel, where Dif is DE (ie, the difference between pixels D and E located at the block boundary).

Table 2 describes the types of blocking effects according to Equation 2.

Output range Type of blocking effect Output value> 11 Blocking effect across 8 pixel areas Output value> 9 Blocking effect across six pixel areas Output value> 7 Blocking effect across four pixel areas Output> 5 Blocking effect occurs over two pixel areas in the left (or upper) block and one pixel area in the right (or lower) block from the boundary of the block Output> 3 Blocking effect occurs over two pixel areas in the right (or lower) block and one pixel in the left (or upper) block from the boundary of the block Output> 1 Blocking effect across two pixel areas Output value ≤ 1 No blocking effect

According to the filtering equation represented by Equation 2, no filtering is performed if no blocking effect has occurred, while the intensity of the smoothing (number of pixels filtered) increases when there is a significant amount of blocking. You can see that.

4 is a flowchart illustrating a process of performing a preferred embodiment of the method for post-filtering a DCT-based image according to the present invention.

Referring to FIG. 4, the block detection detector 110 generates a first input variable I ₁ that is a difference in intensity of blocking based on changes in pixels on both sides of a block boundary of the decoded image (S400). Next, the block detection detector 110 generates a _second and third input variables I ₂ and I ₃ by calculating gradients of pixels in adjacent blocks (S410). Next, the block detection detector 110 determines a membership function corresponding to the first to third input variables based on the relationship between the input variable and the membership function shown in FIG. 3A, and selects an appropriate filter according to the decision rules shown in Table 1 below. Select (S420). Next, the interpolator 120 performs different types of filtering having a filter equation as shown in Equation 2 according to the type of the output membership function that is block information for each block (S430).

In order to obtain objective results such as Peak Signal to Noise Ratio (PSNR) and subjective results such as visual quality, the post-filtering method according to the present invention is referred to as " de-blocking filter for low bit rate MPEG-4 video (de-). blocking removal technique proposed in the article titled "blocking filter" (hereinafter referred to as "Prior Art 1") and blocking blocking technique proposed in the paper entitled "Adaptive Blocking Filter" (hereinafter referred to as "Prior Art 2"). '). Lena, Barbara, Peppers, Goldhill and Boats images of 512 × 512 (pixel) size were used for comparison. Tables 2 to 6 describe the basic JPEG PSNRs for each image by conventional techniques and post filtering techniques according to the present invention. The quality parameter (QP) starts from zero, meaning maximum compression.

QP
(LENA) Bit rate (bpp) JPEG
PSNR (dB) Prior Art 1
PSNR (dB) Prior Art 2
PSNR (dB) The present invention
PSNR (dB) 0 0.1335 24.25 25.53 25.52 25.73 5 0.1732 27.33 28.31 28.30 28.53 10 0.2455 30.41 30.99 31.08 31.29 15 0.3079 31.95 32.30 32.43 32.58 20 0.3634 32.96 33.18 33.33 33.40 25 0.4144 33.70 33.83 33.99 34.00 30 0.4626 34.28 34.34 34.51 34.45

QP
(BARBARA) Bit rate (bpp) JPEG
PSNR (dB) Prior Art 1
PSNR (dB) Prior Art 2
PSNR (dB) The present invention
PSNR (dB) 0 0.1693 22.38 23.11 23.16 23.33 5 0.2113 24.05 24.40 24.53 24.63 10 0.3155 25.79 25.95 26.01 26.19 15 0.4159 27.16 27.25 27.30 27.46 20 0.5080 28.34 28.39 28.46 27.54 25 0.5902 29.39 29.41 29.50 29.42 30 0.6641 30.32 30.34 30.42 30.18

QP
(PEPPERS) Bit rate (bpp) JPEG
PSNR (dB) Prior Art 1
PSNR (dB) Prior Art 2
PSNR (dB) The present invention
PSNR (dB) 0 0.1367 24.29 25.68 25.73 25.91 5 0.1766 27.16 28.21 28.33 28.39 10 0.2463 30.13 30.77 30.89 30.90 15 0.3067 31.53 31.94 32.06 32.02 20 0.3604 32.43 32.69 32.83 32.76 25 0.4143 33.05 33.20 33.37 33.25 30 0.4620 33.53 33.62 33.79 33.64

QP
(GOLDHILL) Bit rate (bpp) JPEG
PSNR (dB) Prior Art 1
PSNR (dB) Prior Art 2
PSNR (dB) The present invention
PSNR (dB) 0 0.1255 23.74 24.79 24.80 24.95 5 0.1710 26.16 26.99 27.00 27.05 10 0.2655 28.65 29.08 29.12 29.20 15 0.3598 29.95 30.20 30.26 30.36 20 0.4448 30.87 31.01 31.10 31.17 25 0.5218 31.56 31.62 31.74 31.78 30 0.5904 32.10 32.12 32.24 32.24

QP
(BOATS) Bit rate (bpp) JPEG
PSNR (dB) Prior Art 1
PSNR (dB) Prior Art 2
PSNR (dB) The present invention
PSNR (dB) 0 0.1349 23.34 24.23 24.20 24.41 5 0.1838 26.03 26.73 26.70 26.91 10 0.2772 28.86 29.30 29.36 29.49 15 0.3574 30.36 30.63 30.71 30.79 20 0.4269 31.41 31.60 31.71 31.70 25 0.4904 32.20 32.33 32.45 32.37 30 0.5480 32.84 32.91 33.05 32.90

The deblocking methods generally provide good results down to 0.5 bit per pixel (bpb), so PSNRs for QPs up to 30 are described in Tables 3-7. According to these results, the post filtering method according to the present invention can improve the performance of JPEG compressed images by 1.5 dB for low bit rates while the gain is limited for higher bit rates. In addition, the post filtering method according to the present invention provides better results when compared with the conventional deblocking methods.

5A to 5E show original images, JPEG compressed images, deblocking filtered images according to the prior art 1, deblocking filtered images according to the prior art 2, and deblocking filtered images by the post filtering method according to the present invention. Five partial Barbara images, respectively, are shown. 5A-5E, the compressed image contains severe blocking. The prior art 1 and the prior art 2 can reduce the blocking effect in the smooth region, but visual blocking occurs in the region where the pattern exists. In contrast, the post filtering method according to the present invention can effectively remove blocking defects without producing any noticeable blurring effect.

The invention can also be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of the computer-readable recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like, and may be implemented in the form of a carrier wave (for example, transmission via the Internet) . The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

Although the preferred embodiments of the present invention have been shown and described above, the present invention is not limited to the specific preferred embodiments described above, and the present invention belongs to the present invention without departing from the gist of the present invention as claimed in the claims. Various modifications can be made by those skilled in the art, and such changes are within the scope of the claims.

1 is a block diagram showing the configuration of a preferred embodiment of the apparatus for post-filtering a DCT based image according to the present invention;

2 shows pixels of a block boundary,

3A and 3B are graphs showing membership functions (MFs) of input and output parameters, respectively;

4 is a flowchart illustrating a process of performing a preferred embodiment of the method for post-filtering a DCT-based image according to the present invention;

5A to 5E are original images, JPEG compressed images, deblocking filtered images according to the prior art 1, deblocking filtered images according to the prior art 2, and deblocking filtered images by the post filtering method according to the present invention. Five partial Barbara images respectively corresponding to FIG.

Claims (7)

Select a predetermined number of pixels in each block sequentially from the boundary of the adjacent blocks in a direction perpendicular to the boundary of the adjacent blocks of the decoded image, and belong to different blocks among the selected pixels, An output for reducing the blocking effect of the blocks by a predetermined decision rule based on the input membership function selected for the second and third input variables, the change values of pixels belonging to the same block as the difference between the first input variable, respectively; Blocking detection unit for determining the membership function; And And an interpolation unit for removing a blocking effect by interpolating pixel values of the selected pixels by a linear filter corresponding to the determined output membership function among linear filters corresponding to each of a preset number of output membership functions. The blocking detection unit is configured to divide the range between a lower limit and an upper limit of values that the first input variable, the second input variable, and the third input variable may have, and thus, the first input variable and the second. And a rating corresponding to a value of each of the input variable and the third input variable to determine the input membership function. The method of claim 1, A plurality of grades set by dividing a range between a lower limit and an upper limit of values that the first input variable, the second input variable, and the third input variable may have are 'low' and 'medium' according to the size of the input variable. 'Low', 'medium high', and 'high', and the output membership function determined by the blocking detector is composed of seven defined by different filter equations. Device. The method according to claim 1 or 2, A post-filtering apparatus for a DCT-based image, wherein the filter equation of the linear filters corresponding to the output membership function is defined as follows:

Where A, B, C, and D are pixel values of pixels sequentially selected in one adjacent block, and E, F, G, and H are pixel values of pixels sequentially selected in another adjacent block, and A ', B', C 'and D' are interpolated pixel values of pixels sequentially selected in one adjacent block, respectively, and E ', F', G 'and H' are selected sequentially in another adjacent block, respectively. The interpolated pixel value of the pixel, where Dif is DE (ie, the difference between pixels D and E located at the block boundary). Select a predetermined number of pixels in each block sequentially from the boundary of the adjacent blocks in a direction perpendicular to the boundary of the adjacent blocks of the decoded image, and belong to different blocks among the selected pixels, Generating second and third input variables that are change values of pixels belonging to the same block as the first input variable, which is a difference value, respectively; The first input variable, the second input variable, and the second input variable from among a plurality of grades set by dividing a range between a lower limit and an upper limit of values of the first input variable, the second input variable, and the third input variable. A grade corresponding to each value of the third input variable is assigned to determine the input membership function, and a blocking effect of the blocks is reduced by a predetermined decision rule based on the input membership functions corresponding to the respective input variables. Determining an output membership function to enable; And And removing the blocking effect by interpolating pixel values of the selected pixels by a linear filter corresponding to the determined output membership function among the linear filters corresponding to each of the preset number of output membership functions. Post filtering method for DCT based images. The method of claim 4, wherein A plurality of grades set by dividing a range between a lower limit and an upper limit of values that the first input variable, the second input variable, and the third input variable may have are 'low' and 'medium' according to the size of the input variable. Low ',' medium high ', and' high ', wherein the determined output membership function is comprised of seven defined by different filter equations. The method according to claim 4 or 5, The filter equation of the linear filters corresponding to the output membership function is defined as follows.

Where A, B, C, and D are pixel values of pixels sequentially selected in one adjacent block, and E, F, G, and H are pixel values of pixels sequentially selected in another adjacent block, and A ', B', C 'and D' are interpolated pixel values of pixels sequentially selected in one adjacent block, respectively, and E ', F', G 'and H' are selected sequentially in another adjacent block, respectively. The interpolated pixel value of the pixel, where Dif is DE (ie, the difference between pixels D and E located at the block boundary). A computer-readable recording medium having recorded thereon a program for executing a post filtering method of a DCT-based image according to claim 4 or 5.

Images