KR101737985B1 - Image processing method - Google Patents

Abstract

The present invention relates to an image processing method capable of providing low complexity and a high deblocking capability for a DCT coded image. The image processing method according to the present invention comprises the steps of: extracting a pixel vector from an image; calculating a reference value of the pixel vector; calculating an activity level indicating a degree of local variation of the pixel vector when the reference value is less than or equal to a predetermined first threshold value; classifying a low-activity model (LAM), a moderate-activity model (MAM), and a high-activity model (HAM) according to the calculated activity level; and applying a preset flat edge filter or a model filter for each model type according to the classified models.

Description

[0001] IMAGE PROCESSING METHOD [0002]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method, and more particularly, to an image processing method that shows high deblocking capability with low complexity.

Conventionally, most image data coding methods process image frames on a block-by-block basis. In this block-based image data encoding method, a discrete cosine transform (DCT) plays a key role. This is because DCT can be a good compromise considering the complexity of computation and the compression capacity of information. Such a DCT is widely used and adopted in coding standards for image data such as JPEG, MPEG, and H.26x.

According to most standards, an image frame is decomposed into a plurality of blocks having N x N sizes. The decomposed blocks are transformed from the spatial domain into the frequency domain through the DCT, and the coefficients obtained through the DCT are quantized. For example, assuming that the size of one decomposed block is 8 x 8, the pixel value of this block is transformed into an 8 x 8 block having one DC coefficient and 63 AC coefficients by DCT, And quantized using a defined quantization table to have a discrete value.

In the block - based coding process using DCT, the blocks are treated as one independent entity and the spatial correlation between adjacent blocks is not considered in the compression coding process. As a result, a deterioration phenomenon appears in the compression coding process of the block unit. That is, the boundary of the blocks in the image reconstructed by decoding the data compressed and coded on a block basis becomes prominent. For example, when a luminance smoothly changes at an adjacent block boundary, the boundary between the blocks becomes conspicuous due to a different quantization interval in the decoding process. This phenomenon is called blocking artifact.

Various methods have been studied to overcome the drawbacks of the DCT based on the block unit, i.e., the block defects. Classification of these methods is roughly divided into two.

One is to use a different encoding scheme, for example, interleaved block coding, lapped transform, combined transform, and the like.

The other is post-processing of the reconstructed image. This method has the advantage that it does not need to change existing standards. There are generally three types of post-processing methods.

The first is to use a low-pass filter to eliminate block faults. The second is to eliminate block faults based on statistical methods, which use a statistical model to correct the DCT coefficients. The third method uses the set theory, which is based on the POCS theory. The POCS theory defines a convex set from a set of image data and constraints, and a projection to the set, and then removes block defects without impairing the original image characteristics through iterative projection to each set.

The post-processing methods described above are required to preserve the boundaries of objects in common and to maintain the sharpness of the image. It is also required that the post-processing method should be simple in order to be applied to a real-time image processing application program.

However, in the first method, over-blurring occurs not only at the boundaries of the blocks but also at the entire image including the boundaries of the objects, and the remaining methods have a problem of complexity of operation, There is a problem that it is difficult to apply to.

Korean Patent Publication No. 10-2005-0099256 (published on October 13, 2005)

It is an object of the present invention to provide an image processing method capable of providing low complexity and high deblocking capability for DCT coded images.

In the image processing method according to the present invention for solving the above problems,

A pixel vector extraction step of extracting a pixel vector from an image; A reference value calculating step of calculating a reference value of the pixel vector; An activity degree calculating step of calculating an activity degree indicating a degree of local variation of the pixel vector when the reference value is equal to or less than a predetermined first threshold value; A low-activity model (LAM) when the calculated activity level is lower than a predetermined second threshold value, a Moderate-activity model (MAM) when the calculated activity level is higher than the predetermined second threshold value and lower than a predetermined third threshold value, And a HAM (High-activity model) when the value is larger than the threshold value 3; Compares the reference value with any one of the predetermined fourth threshold value and the fifth threshold value according to the classification model, and applies a predetermined flat edge filter or a model filter according to each model type Wherein the reference value comprises at least one of

, And the pixel vector

Lt;

May be the number of pixels in a single pixel vector for one deblocking process.

Further, in the activity degree calculating step, the activity degree is calculated by Equation (4)

Is an indicator function,

The

Lt; / RTI >

May be a threshold for determining the relationship between adjacent pixels.

&Quot; (4) "

Further, the first threshold is calculated by Equation (3)

The

From the pixel vector,

= 187.5,

= 0.02,

= -0.009,

Is the standard deviation of the pixel vector, and QF may be a quantization factor.

&Quot; (3) "

Further, in the model classification step, the second threshold value and the third threshold value may be set to 1 and 3, respectively.

When the image is classified as LAM, the filtering step applies the flat edge filter when the reference value is larger than the fourth threshold value, and the fourth threshold value is larger than the first threshold value 0.8, and the flat-edge filter may be expressed by Equation (7) below.

&Quot; (7) "

here,

.

When the image is classified as LAM, the filtering step applies a LAM model filter when the reference value is greater than the fifth threshold and less than the fourth threshold, 1 < / RTI > threshold, the fifth threshold is 0.012 times the first threshold, and the LAM model filter may be: " (6) "

&Quot; (6) "

here,

.

When the image is classified as MAM, the filtering step applies the flat edge filter when the reference value is larger than the fourth threshold value, and the fourth threshold value is larger than the first threshold value 0.6, and the flat-edge filter can be expressed by Equation (7) below.

&Quot; (7) "

here,

.

When the image is classified as MAM, the filtering step applies an MAM model filter when the reference value is greater than the fifth threshold and less than the fourth threshold, 1 < / RTI > threshold, the fifth threshold is 0.012 times the first threshold, and the MAM model filter may be: " (8) "

&Quot; (8) "

When the image is classified as HAM, the filtering step uses the flat edge filter when the reference value is larger than the fourth threshold value, and the fourth threshold value is 0.4 And the flat-edge filter can be expressed by the following equation (7).

&Quot; (7) "

here,

.

In addition, when the image is classified as HAM, the filtering step applies a HAM model filter when the reference value is greater than a fifth threshold value and less than the fourth threshold value, 0.4 times the threshold, the fifth threshold is 0.012 times the first threshold,

Is 0.2 times the first threshold value, and the HAM model filter can be expressed by the following Equation (9) and Equation (10).

&Quot; (9) "

&Quot; (10) "

The present invention provides a method for providing low complexity and high deblocking capabilities for DCT coded images to minimize blocking artifacts and ensure sufficient visual quality of the compressed image.

BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.
Fig. 1 shows the distribution of pixel vectors when K = 8.
2 is a flowchart illustrating an image processing method according to an embodiment of the present invention.
Figure 3 is a graph of the optimal first threshold and QF.
Figure 4 is a graph of the first threshold in relation to the standard deviation of the QF and pixel vector.
Figure 5 shows an edge detection image for a portion of the Lena image.
Figure 6 shows three model categories based on activity.
Figure 7 shows three cases of LAM. (b) Case 2 - Pixel vector with no change in blocking artifacts and one change; (c) Blocking artifacts and flat pixel vectors with no change;
Figure 8 shows the edges of the Lena image.
Fig. 9 shows two cases of MAM when the activity level is 2. Fig. (a) Case 1 - Pixel vector with blocking artifacts, (b) Case 2 - Pixel vector without blocking artifacts
10 shows a 3x3 mask for a 2D directional filter.

BRIEF DESCRIPTION OF THE DRAWINGS The present invention is capable of various modifications and various embodiments, and specific embodiments will be described in detail below with reference to the accompanying drawings.

The following examples are provided to aid in a comprehensive understanding of the methods, apparatus, and / or systems described herein. However, this is merely an example and the present invention is not limited thereto.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail. The following terms are defined in consideration of the functions of the present invention, and may be changed according to the intention or custom of the user, the operator, and the like. Therefore, the definition should be based on the contents throughout this specification. The terms used in the detailed description are intended only to describe embodiments of the invention and should in no way be limiting. Unless specifically stated otherwise, the singular form of a term includes plural forms of meaning. In this description, the expressions "comprising" or "comprising" are intended to indicate certain features, numbers, steps, operations, elements, parts or combinations thereof, Should not be construed to preclude the presence or possibility of other features, numbers, steps, operations, elements, portions or combinations thereof.

It is also to be understood that the terms first, second, etc. may be used to describe various components, but the components are not limited by the terms, and the terms may be used to distinguish one component from another .

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

-Introduce

The block-based discrete cosine transform (BDCT) is widely used for image and video compression, and has several international coding schemes such as JPEG for still images and MPEG [1; 2; 3] for moving images. Standard. BDCT is widely used because of its energy compression nature, which can be said to be almost optimal. It is also a fast algorithm for both image and video compression. At low bit-rates, BDCT-coded images are blocked by blocking artifacts caused by high-frequency transform loss factors during the independent quantization of each block, Blocking Artifacts "). The purpose of the deblocking algorithm is to minimize blocking artifacts and ensure sufficient visual quality of the compressed image. Deblocking methods can be classified into two large approaches (in-loop processing and postprocessing). The in-loop processing has been adopted in H.264 / AVC and can improve the coding efficiency by preventing the expansion of the blocking artifacts between adjacent frames. The post-processing is performed after the images are decompressed. Therefore, post-processing is more practical to remove the blocking artifacts. Many post-processing methods based on deblocking algorithms such as filtering approaches and projection to convex sets (POCS) have been proposed to eliminate deblocking artifacts. The simplest approach is to apply a low-pass filter near the block boundaries. By mitigating the discontinuity, the low-pass filter can achieve superior performance when compared to other techniques. However, the disadvantage of the spatial filtering approach is that the images are smoothed or over smoothed because of its low-pass characteristics.

Numerous methods have been proposed to reduce the blocking artifacts in the DCT domain. Luo et al. Proposed a simple DCT-based deblocking method for smooth regions. In this way, non-smooth areas are not adjusted. This is because the human eye is more sensitive to low frequency signals than high frequency signals. However, this method does not consider the block boundary effect, so it shows low performance. A signal decomposition-based method has been proposed. This method causes over-blurring (and blurring) due to filtering in the DCT domain, especially in the detailed regions. Also, all of these are more complex than spatial filtering.

An iterative postprocessing approach was first proposed by Youla and Webb, who reconstructed the decoded image by iterative projection on several condition sets. These sets are derived from the knowledge of the original image based on the transmitted data and POCS theory. One of the strengths of POCS-based methods is that they are flexible with respect to utilizing known conditions of images not known to the reconstruction. However, Zakhor's method excessively blurs the details of the image. On the other hand, methods such as Yang et al. Are not effective in reducing blocking artifacts; These problems are caused by the repetitive application of an overall low-pass filter and implementation to the pixels belonging to the boundary. Zou's method is more complex, but better performance.

In order to avoid the above mentioned weaknesses, the present invention proposes a new adaptive post-processing deblocking algorithm that shows high deblocking capability with low complexity for DCT coded images. For the first time, the present invention detects sharp edges by adaptive image thresholding that takes into account both image degradation and local image characteristics. Next, the present invention uses a simple criterion for model classification. In order to effectively remove stair artifacts and corner outliers, the present invention implements multiple conditions and is suitable for many block types by thresholds and activities that are specified by local statistical properties Apply a filter. Finally, the present invention uses a simple directional filter to reduce rigning artifacts at edges while retaining image information in a two-dimensional (2-D) direction.

- proposed algorithm

SUMMARY OF THE INVENTION [

Image of size

end

Vector. By raster-scanning from rows to lexicographical order, for example,

And

.

Is the pixel value of each pixel in the image. Next,

(PV) < / RTI >

. From here

,

And

The

Can not be the same as a multiple of.

Is the number of pixels in a single pixel vector for one deblocking process. In addition, the present invention assumes that the blocking artifacts are symmetric. Therefore, based on this implicit assumption

Is an even number

to be. The block size of the BDCT is , And the pixel vector is

Lt; RTI ID = 0.0 > block boundaries. Image Width

And height

Is the length of the pixel vector

. Figure 1 depicts pixel vector extraction from block boundaries. Pixel vectors are processed one by one in columns or rows. The horizontal blocking artifacts are reduced using the same method used for vertical. When the present invention processes it horizontally, the present invention only rotates the image by 90 degrees and processes it in the same way as the vertical deblocking method.

2 is a flowchart illustrating an image processing method according to an embodiment of the present invention.

An image processing method according to the present invention includes: extracting a pixel vector from an image (S210); A reference value calculating step (S220) of calculating a reference value of the pixel vector; An activity degree calculating step (S230) of calculating an activity degree indicating a degree of local variation of the pixel vector when the reference value is smaller than a predetermined first threshold value; A low-activity model (LAM) when the calculated activity is less than a predetermined second threshold, a Moderate-activity model (MAM) when the calculated activity is greater than the predetermined second threshold and smaller than a predetermined third threshold, A high-activity model (HAM) if the first threshold value is greater than a third threshold value of the second threshold value (S240); Compares the reference value with any one or more of the predetermined fourth threshold value and the fifth threshold value according to the classification model, and applies a flat edge filter or a model filter according to each model type according to the comparison. And a filtering step (S250). If the reference value is greater than the first threshold value in step S230, a directional filter may be separately applied in step S250. Each step will be described in more detail below.

- Image binarization Thresholding ) produce

Image binarization (Thresholding) is used to capture discontinuities between block boundaries. ; The choice of the threshold is an important factor in applying the algorithm to find the pixel vector containing the blocking artifact. In the present invention, local image binarization (Thresholding) is designed as follows.

From here

Is a first threshold used to identify an edge of an image, and a notation,

The

PV. Also,

Is referred to as a reference value. The deblocking algorithm must be able to avoid over-smoothing, as well as be able to distinguish edges in order to preserve the sharp edges of the actual image within the block boundary. The simplest way to accomplish this is to set it to a constant value. However, each discontinuity at the block boundary is different due to the effect of degradation. For example, there is a strong correlation between the threshold and the DC quantization factor (QF). Blocking artifacts are more severe in large QFs. The threshold is used to separate the edges from the blocking artifact. Therefore, in the present invention, the larger the QF, the larger the threshold value. ; An increase in the threshold is associated with an increase in QF. The first threshold

(Below,

) Is defined as Equation (2) using QF.

From here

Wow

Is an empirically derived parameter,

= 187.5

= 0.02. The optimal threshold is shown in FIG.

Given the fact that local characteristics can change rapidly in the image, it is necessary to adjust the threshold using local image characteristics. therefore

Depends not only on QF but also on local characteristics. Pixel vectors are processed using different methods because of their different properties. For smooth regions, the threshold must be large enough to effectively remove any blocking artifacts. Also, when the variant of the pixel vector is smooth,

Extracts and protects edge information across the block boundary, and removes block discontinuities in the flat pixel vector. On the other hand, the threshold for a complex region must be small in order to prevent image details from blurring. Because standard deviations are a measure of the amount of change in values, they are often used as an indicator of image variation, and threshold definitions include standard deviations. The standard deviation has a small value in the smooth areas and a large value in the complicated areas. The threshold is therefore associated with a reduction in the standard deviation. In the present invention,

Can be modified.

From here

Is the standard deviation of the pixel vector,

= -0.009. 3,

And QF and

Are shown. Figure 4

Figure 2 shows an example of edge detection.

In the present invention, three models are classified according to the degree of flatness of each pixel vector. : a low-activity model (LAM), a moderate-activity model (MAM), and a high-activity model (HAM). In order to reflect the actual properties of the pixel vector, the present invention designs a criterion for classifying the type of the pixel vector. Accordingly, in the present invention, in order to capture a relation between adjacent pixels,

. Also, the activity of the pixel vector

The type of the pixel vector. The activity level is defined by Equation (4).

From here

Is an indicator function,

The set

Lt; / RTI >

Is a threshold for determining the relationship between adjacent pixels. once

The second threshold value < RTI ID = 0.0 >

(Below,

) And the third threshold value

(Below,

).

The model decision rules are determined as follows. : if

end

, The pixel vector belongs to the LAM, and if

end

The pixel vector belongs to the HAM, and if

end

Bigger

Or less, the pixel vector belongs to the MAM. Parameters

,

And

Can be determined empirically, for example,

= 1,

= 3 and

= 3.

- Filtering  process

In the present invention, filters are designed to be applicable to three models. First, we start with the LAM, the simplest model that is less complex and less varied than the other two models, MAM and HAM. Figure 7 shows three possible cases of pixel motion in the LAM. In the present invention,

Is assumed to be 1 or 0.

Figure 7 (a) shows the blocking artifacts at the block boundary

Shows Case 1 with 1. 7 (b)

Case 1 is shown in Fig. However, there is no change in pixels at the block boundary, which may be a flat edge. Case 3 is a flat pixel vector with no blocking artifacts, and the change is shown in Fig. 7 (c). Case 2 and Case 3 have no blocking artifacts at the block boundary. Therefore, in order to reduce computational complexity, it is not necessary to apply deblocking filters to these areas. In the present invention, to extract a pixel vector without a blocking artifact at the block boundary, another artifact-free fifth threshold

(Below,

Is designed.

Should be adapted with respect to the characteristics of the images. Empirically,

0.012

And is defined by the following equation (5).

If the above condition is satisfied, the pixel vector is considered as artifact-free. Otherwise, the pixel vector is Case 1 with a blocking artifact and is processed with a LAM deblocking filter.

The pixels of each pixel vector in the LAM depend on other pixels. Therefore, the smooth filter should be applied to all pixels in the pixel vector. Otherwise, unexpected new blocking artifacts may occur in unprocessed pixels. On the basis of these issues, the filter is designed in Equation (6).

From here

to be.

However, as shown in Fig. 8, there is a problem that the filter appears clearly at the flat edges of the image. Indeed, the threshold used in the present invention is able to separate sharp contrast edges from images in which pixel values change rapidly. However, in the present invention, this is ignored in consideration of the fact that some flat edges are present, and this can be determined in the LAM category. In this case, the smoothing filters presented above are not suitable. In order to preserve the details of the image and to make the human eye more comfortable, a new adaptive filter is needed,

To be categorized into two types. Flat edge threshold in LAM

Respectively,

= 0.8

to be. A flat edge filter is defined by Equation (7).

here,

to be. Also, the remaining pixels are not filtered. In summary, the following judgment is made to determine which filter to use. : if

, The filter defined by equation (6) is applied, otherwise the filter defined by equation (7) is applied.

The activity of a pixel vector in a MAM is 2 or 3. In the present invention, two cases are assumed for the MAM. Case 1 - PV has a blocking artifact at the block boundary, and Case 2 - PV has no change at the boundary. 9 is a cross-

= 2 is shown. Similar to LAM, Case 2 is artfact-free, and in the present invention,

Lt; / RTI > However, the filter used for Case 1 is more complex than that used in LAM, and is defined as in Equation 8.

The remaining pixels are not filtered.

As described in the LAM, in the present invention, the flat edge threshold of the MAM

Should be used to distinguish the flat edges. MAM has higher activity than LAM, which has a higher probability of including flat edges. In the present invention,

And

= 0.6

. The flat edge filter in the MAM also uses the one defined in Equation (7).

The human eye is not sensitive to the HAM domain. Because the vectors have oscillated pixel values. Therefore, it becomes unnecessary to adjust the pixel values of the PV. ; Therefore, in the present invention, two pixels can be modified to avoid over-smoothing at block boundaries. For this purpose, in the present invention, three threshold values

,

And

. ≪ / RTI > Threshold

Was introduced above,

Is used to distinguish the flat edges using the flat edge threshold of the HAM. Threshold

Is a new threshold and is used to design filters for other local pixel relationships.

Two thresholds (

and

) Is necessary because the design of the filters is based on the relationship between neighboring pixels. However, when two pixels are adjacent, there is a problem as to which pixel should be selected for the filter. Here, simple rules are assumed in the present invention. : If there is a difference between the pixels at the block boundary and another pixel that is not at the block boundary

, The pixel depends on all of the adjacent pixels. Otherwise, the present invention simply adjusts the pixels of the two block boundaries. Therefore, in the present invention, it is necessary to identify the difference between two adjacent pixels, and design filters according to the difference. if

, This is a flat edge and equation (7) is applied. Otherwise, a filter (model filter) for HAM is designed using Equations (9) and (10).

Empirically,

and

Respectively.

And 0.2

Lt; / RTI > In the present invention, the fourth threshold is

,

And

to be. In addition, the model filter of the LAM in the present invention can be defined by Equation (6), the model filter of the MAM may be defined by Equation (8), and the model filter of the HAM may be defined by a combination of Equation (9) and Equation (10), respectively.

- Directional filter

The edge blocks include ringing artifacts that are difficult to remove using the one-dimensional (1D) method. : Ringing artifacts are visually undesirable, although in terms of PSNR they are only the cause of slight deterioration. Since the strong edges can be located well inside the block, it is not possible to suppress ringing artifacts by smoothing along the boundaries of the block using a 1-D filter. A better and simpler way is to create a 2D directional filter that can suppress ringing artifacts while preserving directional information. Therefore, in order to remove ringing artifacts, a 2D direction filter is applied to a 3 × 3 mask as shown in FIG. The pixels to be adjusted

Quot; In the present invention,

Is obtained by using Equation (11).

here,

Is a weight function defined in Equation (12). And

The

A point is the set of coordinates of a 3 x 3 window at its center.

here,

to be.

So the amount of calculation is halved. The pair of pixels is averaged in the direction with the highest associated, and this filtering process can effectively reduce artifacts while preserving the original edges. Also, in the present invention, a normalized weight function is selected,

.

-conclusion

First, the present invention introduces an adaptive threshold based on a quantization factor (QF) and a local statistical characteristic, and the adaptive threshold separates edges and blocking artifacts . Next, the activity of the pixel vector (PV) is used as an index to classify the pixel vector into three model types. In the present invention, corresponding filters are created based on the characteristics of each model type. In the edge region, the deblocking filter is not applied to prevent over-smoothing. Instead, the present invention introduces a directional filter to reduce ringing artifacts and improve edge pixels.

The foregoing description is merely illustrative of the technical idea of the present invention, and various changes and modifications may be made by those skilled in the art without departing from the essential characteristics of the present invention. Therefore, the embodiments described in the present invention are not intended to limit the technical spirit of the present invention but to illustrate the present invention. The scope of protection of the present invention should be construed by the following claims, and all technical ideas within the scope of equivalents thereof should be construed as being included in the scope of the present invention.

Claims (11)

A pixel vector extraction step of extracting a pixel vector from an image;
A reference value calculating step of calculating a reference value of the pixel vector;
An activity degree calculating step of calculating an activity degree indicating a degree of local variation of the pixel vector when the reference value is equal to or less than a predetermined first threshold value;
A low-activity model (LAM) when the calculated activity level is lower than a predetermined second threshold value, a Moderate-activity model (MAM) when the calculated activity level is higher than the predetermined second threshold value and lower than a predetermined third threshold value, And a HAM (High-activity model) when the value is larger than the threshold value 3;
Compares the reference value with any one of the predetermined fourth threshold value and the fifth threshold value according to the classification model, and applies a predetermined flat edge filter or a model filter according to each model type And a filtering step,
The reference value

, And the pixel vector

Lt;

Is the number of pixels in a single pixel vector for one deblocking process,
In the activity degree calculating step,
The activity level is calculated by Equation (4)

Is an indicator function,

The

Lt; / RTI >

Is a threshold for determining a relationship between adjacent pixels,
&Quot; (4) "

,
The first threshold is calculated by Equation (3)

The

From the pixel vector,

= 187.5,

= 0.02,

= -0.009,

Is the standard deviation of the pixel vector, QF is the quantization factor,
&Quot; (3) "

,
If the image is classified as LAM,
Wherein the filtering step uses the flat edge filter when the reference value is larger than the fourth threshold value,
Wherein the fourth threshold is 0.8 times the first threshold,
The flat-edge filter is expressed by Equation (7) below,
&Quot; (7) "

here,

,
If the image is classified as LAM,
Wherein the filtering step applies a LAM model filter when the reference value is greater than the fifth threshold and less than the fourth threshold,
The fourth threshold is 0.8 times the first threshold, the fifth threshold is 0.012 times the first threshold,
The LAM model filter is expressed by Equation (6) below,
&Quot; (6) "

here,

Lt;
Wherein when the reference value is equal to or less than the fifth threshold value, the deblocking filter is not applied, considering that there is no blocking artifact at the block boundary.
delete delete The method according to claim 1,
In the model classification step,
Wherein the second threshold value and the third threshold value are set to 1 and 3, respectively.
delete delete The method according to claim 1,
If the image is classified as MAM,
Wherein the filtering step uses the flat edge filter when the reference value is larger than the fourth threshold value,
The fourth threshold is 0.6 times the first threshold,
Wherein the flat-edge filter is expressed by the following equation (7).
&Quot; (7) "

here,

.
The method according to claim 1,
If the image is classified as MAM,
Wherein the filtering step applies an MAM model filter when the reference value is greater than the fifth threshold and less than the fourth threshold,
The fourth threshold is 0.6 times the first threshold, the fifth threshold is 0.012 times the first threshold,
Wherein the MAM model filter is the following expression (8).
&Quot; (8) "

.
The method according to claim 1,
If the image is classified as HAM,
Wherein the filtering step uses the flat edge filter when the reference value is greater than a fourth threshold value,
The fourth threshold is 0.4 times the first threshold,
Wherein the flat-edge filter is expressed by the following equation (7).
&Quot; (7) "

here,

.
The method according to claim 1,
If the image is classified as HAM,
The filtering may be performed by applying a HAM model filter when the reference value is greater than a fifth threshold and less than or equal to the fourth threshold,
The fourth threshold is 0.4 times the first threshold, the fifth threshold is 0.012 times the first threshold,

Is 0.2 times the first threshold,
Wherein the HAM model filter is expressed by the following equations (9) and (10).
&Quot; (9) "

&Quot; (10) "

.
A program stored in a computer-readable medium for executing the steps of any one of claims 1, 4 and 7 to 10 in a computer.

Images