KR100711725B1 - Method for filtering of deblocking in video telephony mobile phone - Google Patents

Abstract

The present invention relates to a deblocking filtering method of a video telephone, and provides a deblocking filtering method of a video telephone that can perform efficient filtering suitable for a very limited system such as a video telephone by simplifying an operation structure. The present invention for this purpose is an extraction step of extracting three adjacent pixels of each block around the block boundary; Calculating a difference δ according to the distribution of the pixels with respect to the block boundary; A mode determining step of determining a mode for performing filtering according to a degree of change (SF) between adjacent pixels of the pixels; A determination step of determining whether to perform filtering according to the relation between the difference δ and a preset quantization parameter QP; And a filtering step of determining a target pixel to be filtered according to the mode determination and the determination result, and performing a different level of filtering for each. With the above configuration, the present invention can simplify the computational structure for deblocking filtering to reduce computational complexity and provide subjective picture quality improvement. Therefore, the present invention can be used at low bit rates such as H.263 and MPEG-4. There is an effect that can be applied to the video encoding method.

Videophone, deblocking, maximum distortion limit model, delta, MPEG-4, H.263

Description

Method for filtering of deblocking in video telephony mobile phone}

1 is a flowchart illustrating a deblocking filtering method of a video telephone according to an embodiment of the present invention.

Figure 2 is a schematic diagram showing a block boundary for applying the deblocking filtering method of a video telephone according to an embodiment of the present invention.

FIG. 3 is a block diagram illustrating the delta operation of FIG. 1. FIG.

4 is a flowchart illustrating a mode determination method of FIG. 1.

5 is a flow chart illustrating a filtering method by the maximum distortion limit model of FIG. 1.

6 is a block diagram illustrating the lowpass filtering of FIG.

The present invention relates to a deblocking filtering method of a video telephone, and more particularly, to a deblocking filtering method for improving the image quality of a video telephone, wherein the target region to be filtered is adapted according to distribution characteristics of pixels. The present invention relates to a deblocking filtering method of a video telephone that can be determined more efficiently and to apply different levels of filtering in each case to reduce the computational complexity for deblocking.

In recent years, mobile communication terminals have tended to expand their application ranges from conventional voice calls to multimedia functions such as video.

Accordingly, video encoding is used to efficiently process a large amount of image data. In this video encoding, blocking artifacts occur at a block boundary of a reconstructed image due to quantization, which is perceived as a deterioration of image quality. do.

On the other hand, current video telephones perform encoding at a target bit rate as low as 48 kbps, so that image quality deteriorates severely. Therefore, providing improved picture quality to end-users has become a very important issue.

In addition, in video telephones, video encoding / decoding and audio encoding / decoding must be performed together with other UI and communication protocols.

In particular, Qualcomm's MSM6XXX chipset, which is currently in use, leaves only a limited amount of spare resources available to improve videophones.

Therefore, the most important part of the video phone is the improvement of subjective quality and the simplification of operation.

The proposed methods in the existing standard propose post-processing deblocking filter in MPEG-4 Annex-J, which is not suitable for systems requiring low computational amount because it requires large computational amount.

In addition, the method proposed in H.263 has a faster and simpler amount of computation than the MPEG-4 method, but has a problem in terms of image quality that is lower than that of the MPEG-4 Annex-J method.

The present invention has been proposed to solve the above problems, and it is possible to perform efficient filtering on a very limited system such as a videophone by simplifying the computational structure by using the difference between adjacent pixels and the maximum distortion limit model around the block boundary. An object of the present invention is to provide a deblocking filtering method of a video telephone.

The present invention for achieving the above object is an extraction step of extracting three adjacent pixels of each block around the block boundary; Calculating a difference δ according to the distribution of the pixels with respect to the block boundary; A mode determining step of determining a mode for performing filtering according to a degree of change (SF) between adjacent pixels of the pixels; A determination step of determining whether to perform filtering according to the relation between the difference δ and a preset quantization parameter QP; And a filtering step of determining a target pixel to be filtered according to the mode determination and the determination result, and performing a different level of filtering for each.

Preferably, the mode determining step calculates the number of cases where the difference value between adjacent pixels of the pixels exceeds a preset threshold value TH as the degree of change SF; The calculated change degree SF is compared with a preset threshold value THU, and when the calculated change degree SF is equal to or greater than the threshold value THU, the control unit judges a soft area and otherwise determines a complex area. It may include a step.

Preferably, the preset threshold THU may be four.

Preferably, the first filtering step of performing offset-based filtering in accordance with the determination result, if the filtering step is determined to be a soft region in the determination step; If it is determined as a complex region in the determination step, it may include a second filtering step for performing low pass filtering according to the determination result.

Preferably, when the first filtering step determines that the absolute value of the difference δ is less than the quantization parameter QP, the same filtering value is applied to pixels at positions symmetric about the boundary of the block. You can add or subtract.

Preferably, the offset value can be calculated using the maximum distortion limit model.

Preferably, the maximum distortion limit model may be calculated by modeling a rate-distortion (R-D) curve obtained through an average value of distortions generated according to the quantization parameter QP.

Preferably, in the second filtering step, when the absolute value of the difference δ is less than 8 times the quantization parameter QP, low pass filtering is performed on two adjacent pixels around the boundary of the block. Can be.

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

In the deblocking filtering, the present invention is to achieve a more subjective improvement in image quality with a low computational level of H.263, and determines a target region to be filtered by varying the filtering conditions according to the distribution characteristics of pixels. This reduces the computational complexity, applies different levels of filtering for each condition, and in particular, minimizes the effects of excessive filtering using the maximum distortion limit model to achieve subjective picture quality improvements while reducing computational complexity. Can be.

Deblocking filtering method of a video telephone according to an embodiment of the present invention, as shown in Figure 1, the extraction step (step S101) of extracting three adjacent pixels of each block around the block boundary, and the block boundary An operation step (step S102) of calculating a difference δ according to the distribution of the pixels for a mode, and a mode determination step (step S103) of determining a mode for performing filtering according to the degree of change SF between adjacent pixels of the pixels. A determination step (steps S104 and S107) of determining whether to perform filtering according to the relationship between the difference δ according to the distribution of the pixels and the preset quantization parameter QP; and filtering according to the mode decision and the determination result. Determining a target pixel to be performed, and performing a different level of filtering for each step (steps S105 and S108), and outputting the result on the screen (steps S109 to S110).

Hereinafter, a method of deblocking filtering of a video telephone according to an embodiment of the present invention will be described in detail with reference to FIGS. 1 to 6.

1 is a flowchart illustrating a deblocking filtering method of a video telephone according to an embodiment of the present invention, and FIG. 2 is a schematic diagram showing a block boundary for applying the deblocking filtering method of a video telephone according to an embodiment of the present invention. .

First, in step S101, when three adjacent pixels of each block are extracted around the block boundary, the deblocking filtering method of the video telephone according to the embodiment of the present invention is disclosed.

That is, as shown in FIG. 2, three pixels A [1] to A [6] are extracted from each block around the boundary of the block as a target region to be filtered.

In step S102, a difference δ according to the distribution of pixels with respect to the block boundary is calculated.

More specifically, the delta (delta, δ), which is the degree of difference in consideration of the distribution characteristic between adjacent pixels with respect to the block boundary, is calculated by the following equation.

는 최대 정수값을 구하는 연산자이다. Here, S2 ~ S5 is the value of the pixel,

Is the operator to get the maximum integer value.

The delta (δ) is used for mode determination for filtering as described below, and an example of implementation of the delta (δ) operation using Equation 1 will be described.

FIG. 3 is a block diagram illustrating the delta operation of FIG. 1.

The delta (δ) operation uses block boundaries A [43], A [4] and each of the pixels adjacent thereto, as shown in FIG. 2, with pixels A [3] located at the block boundary. , A [4]) gives a factor of 4 times, adds the difference between the pixels A [2], A [5] which are located next, and divides the value by 8.

Here, since the main cause of the blocking artifact is due to the difference value of the pixels immediately adjacent to the block boundary, the other part A [2] is used to reflect more influence on the corresponding parts A [3] and A [4]. ], Which is larger than A [5]).

The delta δ calculated as described above is finally subjected to an integerization process, in order to simplify the operation structure by excluding a floating operation in a subsequent filtering operation.

In step S103, a mode for performing filtering is determined according to the degree of change SF between adjacent pixels of the pixels.

That is, in order to determine the filtering mode according to the distribution characteristic, the distribution of values of pixels is analyzed to distinguish whether a corresponding image is a smooth region or a complex region.

Here, the soft region means that the pixels in the target block to be filtered have a relatively uniform value, and the complex region means that the pixels have different values.

On the other hand, since human vision is sensitive to small changes in the soft image area, it is necessary to perform high level strong filtering, and it is insensitive to small changes in the complex image area. Weak Filtering

The different soft levels and complex areas are filtered at different levels to prevent distortion of the image due to excessive blurring in consideration of their characteristics. As such, the range of pixels for which filtering is to be performed is determined.

A more detailed description of the mode determination method will be described later with reference to FIG. 4.

As a result of the determination in step S103, in the case of determining a smooth area, the process proceeds to step S104 where the delta δ is compared with the preset quantization parameter QP, and when the absolute value of the delta δ is smaller than the quantization parameter QP, The flow advances to step S105 to perform filtering by the maximum distortion limit model.

This filtering is offset-based filtering, which adds or subtracts the same offset value for pixels at positions symmetric about the boundary of the block, where the offset value must be considered for each characteristic. As described later, it is calculated using a Maximum Distortion Bound (MDB) model in consideration of a Rate-Distortion (RD) relationship related to the loss generated during filtering.

In this strong filtering performed on the soft region, filtering is performed on a total of six pixels, three pixels each on the left and right sides of the block boundary.

If the absolute value of the delta δ is greater than or equal to the quantization parameter QP, the process proceeds to step S106, skips without performing deblocking filtering, and then proceeds to step S109.

As a result of the determination in step S103, in the case of determining a complex area, the flow advances to step S107 to compare the delta δ with the preset quantization parameter QP, and the absolute value of the delta δ is greater than 8 times the quantization parameter QP. If it is small, the flow advances to step S108 to perform low pass filtering.

The weak filtering performed on the complex region performs filtering only on a total of two pixels, one pixel of each block adjacent to the block boundary, and an implementation thereof is as follows.

FIG. 6 is a block diagram illustrating the low pass filtering of FIG. 1.

Lowpass filtering uses a filter tap of [1 2 1], as shown in FIG. 6, which is performed only for pixels immediately adjacent to the block boundary.

That is, as shown in FIG. 2, the filtering is performed only on the pixel A [3] at the left block boundary and the pixel A [4] at the right block boundary. In this case, adjacent pixels of the pixel to be filtered are performed. I use it.

Here, the low pass filtering using the filter tap of [1 2 1] gives weights to the pixels to be filtered, and the target pixels are weighted twice as much as the adjacent pixels, and the values are averaged.

As a result of the determination in step S107, if the absolute value of the delta δ is greater than or equal to 8 times the quantization parameter QP, the process proceeds to step S106, skips without performing deblocking filtering, and then proceeds to step S109. do.

Steps S101 to S108 are performed on 8 × 8 blocks adjacent to each other, and the same process is repeated for all images.

In step S109, the image is reconstructed according to the filtered pixel value.

That is, after performing the processes of steps S101 to S108 for all images, a new image is constructed using the result, that is, the filtered pixel values.

In step S110, the reconstructed image is displayed.

In this manner, the deblocking filtering method of the videophone according to the embodiment of the present invention is terminated.

4 is a flowchart illustrating a mode determination method of FIG. 1.

First, in step S401, six adjacent pixel values are extracted around the block boundary. Here, the pixel reads the value extracted in step S101.

In step S402, an operation for obtaining the degree of change SF between adjacent pixels of the pixels is started. Here, SF is initialized to 0, variable i is set to 1, and so on.

As described above, the SF (Sample variation) is used to determine the primary mode for determining the filtering level, and the difference value between adjacent pixels of the pixels exceeds the preset threshold TH. It is calculated as a number.

In step S403, it is determined whether the difference value between adjacent pixels exceeds the preset threshold value TH, and when it is determined that the difference value does not exceed, the flow advances to step S404 to increase SF by 1 and proceeds to step S405.

Here, the threshold value TH can be determined by a prior experiment.

If it is determined in step S403 that the difference between adjacent pixels exceeds the preset threshold TH, the process proceeds to step S405 where the variable i is increased by one.

That is, if the difference between adjacent pixels is less than or equal to the threshold value TH, the SF value is increased by 1, otherwise, no change is made to the SF value.

In step S406, if it is determined that the variable (i) exceeds 5 and does not exceed, it returns to step S403 to determine whether the difference value between the next adjacent pixels of steps S403 to S405 exceeds a preset threshold TH. Iteratively performs the operation of SF accordingly.

Here, since the number of extracted pixels is six, five difference values between adjacent pixels are calculated, so that the iterative loop of steps S403 to S406 is repeated until the variable i exceeds 5, that is, only five times.

In step S407, the calculated change degree SF is compared with the preset threshold value THU, and when the calculated change degree SF is equal to or greater than the preset threshold value THU, the flow advances to step S408 to determine a smooth area. do.

Here, the preset threshold THU may be preferably 4, which means that a difference value for four or more sections among the difference values between six adjacent pixels around the block boundary is greater than the preset threshold TH. It is a small case.

Such a soft region means that the pixels have similar values as a whole, and since the human's vision is sensitive to the change, the process returns to step S104 to perform strong filtering.

As a result of the determination in step S407, if the calculated change degree SF is smaller than the preset threshold value THU, the flow advances to step S409 to determine a complicated area.

Such a complicated area means that the pixels have different values as a whole, and since the human's vision is insensitive to the change, the process returns to step S107 to perform weak filtering.

5 is a flowchart illustrating a filtering method using the maximum distortion limit model of FIG. 1.

First, in step S501, the absolute value of delta δ according to the distribution of pixels is compared with the factor Factor [QP] according to the maximum distortion limit model for the corresponding quantization parameter QP.

Here, Factor [] is an array having 31 values, and the values constituting the array are calculated using a maximum distortion limit model.

The maximum distortion limit model is modeled as follows by analyzing a rate-distortion (R-D) curve obtained through an average value of distortions generated according to the quantization parameter (QP).

In addition, the maximum distortion limit model has a problem of speed when the actual values are implemented through calculations, so that the values calculated in advance can be used as a table, and the problem of floating point operations by integerizing the final values. By eliminating, the complexity of the operation can be greatly improved.

는 최대 정수값을 구하는 연산자이다. Where y is a factor value according to the maximum distortion limit model, x is a quantization parameter,

Is the operator to get the maximum integer value.

In the present invention, the maximum distortion limit model is modeled using a Foreman and Carphone sequence having image characteristics similar to those of a video telephone. The factor values are as follows.

Factor [31] = {0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 6, 6}

As a result of the comparison of step S501, if the absolute value of the delta δ is greater than the factor [QP] value, the flow advances to step S502 and the absolute value of the delta δ is 2 of the factor [QP] value. When the absolute value of the delta δ is large, the process proceeds to step S503, where the first offset value d1 is determined to be zero.

As a result of the comparison in step S502, if the absolute value of delta δ is less than or equal to twice the value of factor [QP], the flow advances to step S504 and the delta (δ) at twice the factor Factor [QP]. D1 is calculated by subtracting the absolute value of

As a result of the comparison of step S501, if the absolute value of the delta δ is less than or equal to the factor (QP) value, the flow advances to step S505 to determine the absolute value of the delta δ as d1.

The calculation of the first offset value according to the above steps S501 to S505 can be summarized by the following equation.

Here, y is a factor value according to the maximum distortion limit model.

In step S506, the sign of the delta (δ) is assigned to the value d1 calculated by the expression (3).

That is, since the offset values calculated in steps S503 to S505 are magnitude values by absolute value calculation, the original sign of the delta δ, which is the degree of change between adjacent pixels, is added to finally calculate the first offset value d1. do.

In step S507, filtering is performed on pixels closest to the block boundary, for example, A [3] and A [4], using the calculated first offset value d1 according to the following equation. do.

A [3] = A [3] + d1

A [4] = A [4]-d1

Next, in steps S508 to S510, the second offset value d2 is calculated by the following equation using the first offset value d1.

Here, min () is an operator for selecting a small value, max () is an operator for selecting a large value, and S2 and S5 are values of the corresponding pixel.

That is, in step S508, the first offset value d1 is divided by two (A), and the pixels at the second adjacent symmetrical positions around the block boundary, for example, the second pixel and the fifth pixel in FIG. Divide the absolute value of the difference by 4 by (B).

In step S509, a larger value is selected from the value of A multiplied by -1 and the value of B calculated in step S508 (C). In step S510, the smaller value of the C value and A value is converted to the second offset value d2. Decide

In step S511, the second adjacent symmetric position pixel centered on the block boundary using the calculated second offset value d2, for example, in A [2] and A [5] in FIG. Filtering is performed according to an expression.

A [2] = A [2] + d2

A [5] = A [5]-d2

Next, in steps S512 to S515, the third offset value d3 is calculated by the following equation using the second offset value d2.

Here, min () is an operator that selects a small value, max () is an operator that selects a large value, and S6 and S1 are values of the corresponding pixel.

That is, in step S512, the second offset value d2 is divided by 2 (A), and the pixels of the third adjacent symmetrical position around the block boundary, for example, the first pixel and the sixth pixel in FIG. Divide the absolute value of the difference by 4 by (B).

In step S513, a larger value is selected from the value of A multiplied by -1 and the value of B calculated in step S512 (C). In step S514, the smaller value of the C value and A value is converted to the third offset value d3. Decide

In step S515, the third adjacent symmetric position pixel around the block boundary using the calculated third offset value d3, for example, in A [1] and A [6] in FIG. Filtering is performed according to an expression.

A [1] = A [1] + d3

A [6] = A [6]-d3

As such, when the filtering by the maximum distortion limit model of each of the three pixels adjacent to the block boundary with respect to the soft area is finished, the process returns to step S109 for reconstruction of the image.

On the other hand, in the present invention, since the resources are not provided enough to perform all the block boundaries in performing the filtering, in the present invention, setting the conditions for performing the filtering is very important.

In addition, since the range of available MIPS is further reduced by switching to 18-bit display, in consideration of this, the present invention performs filtering only on intra-coded blocks as in H.263.

As described above, the deblocking filtering method of the video telephone according to the present invention adaptively determines a target region to be filtered according to the distribution characteristics of pixels, and performs different levels of filtering for each case. In addition, the computational structure for deblocking filtering can be simplified to reduce computational complexity and to provide subjective improvement of image quality.

The video is used at low bit rates such as H.263 or MPEG-4 by performing strong filtering on smooth areas using only the pre-tabled values using the maximum distortion limit model and the quantization parameter values that are the most basic video encoding information. Applicable to the coding scheme.

Claims (8)

An extraction step of extracting one pixel adjacent to the block boundary and two pixels adjacent thereto from each block with respect to the block boundary; Calculating a difference δ according to the distribution of the pixels with respect to the block boundary; A mode determining step of determining a mode for performing filtering according to a degree of change (SF) between adjacent pixels of the pixels; A determination step of determining whether to perform filtering according to the relation between the difference δ and a preset quantization parameter QP; And determining a target pixel to be filtered according to the mode determination and the determination result, and performing a different level of filtering for each of the deblocking filtering methods of the video telephone. The method of claim 1, The mode decision step, Calculating the number of cases where the difference value between adjacent pixels of the pixels exceeds a preset threshold value TH as the variation degree SF; The calculated change degree SF is compared with a preset threshold value THU, and when the calculated change degree SF is equal to or greater than the threshold value THU, the control unit judges a soft area and otherwise determines a complex area. Deblocking filtering method of a video telephone, characterized in that it comprises a step. The method of claim 2, And said predetermined threshold value (THU) is four. The method of claim 2, The filtering step, A first filtering step of performing offset-based filtering according to the determination result when the soft region is determined in the determination step; And a second filtering step of performing low pass filtering according to the determination result when the determination is made to the complex region in the determination step. The method of claim 4, wherein In the first filtering step, when the absolute value of the difference δ is smaller than the quantization parameter QP, the first filtering step adds or subtracts the same offset value to pixels at positions symmetric about the boundary of the block. Deblocking filtering method of a video telephone, characterized in that. The method of claim 5, And the offset value is calculated by using a maximum distortion limit model. The method of claim 6, The maximum distortion limit model is calculated by modeling and calculating a rate-distortion (R-D) curve obtained through an average of distortions generated according to the quantization parameter (QP). The method of claim 4, wherein In the second filtering step, when the absolute value of the difference δ is less than 8 times the quantization parameter QP, the low pass filtering of two adjacent pixels around the boundary of the block is performed. Deblocking filtering method.

Images