KR100687845B1 - Filtering method for pixel of image - Google Patents

Abstract

The present invention relates to an image pixel filtering method. In particular, in realizing a compressed video standardization method, it is possible to facilitate real-time processing by filtering pixels of a block boundary region based on at least one of motion vector information, transform coefficient information, and encoding type information. Of course, the purpose of ensuring the quality of the compressed video. According to an aspect of the present invention, there is provided a method of filtering pixels in a block boundary region of an image, the method including: a first process of obtaining motion vector information, transform coefficient information, and encoding type information of a block in contact with a block boundary; And to filter the pixels in the block boundary area based on a filtering method including a second process of determining whether to perform filtering on the block boundary area based on at least one of the obtained information.

Description

How to filter image pixels {FILTERING METHOD FOR PIXEL OF IMAGE}

1 is an exemplary view showing a block position in the horizontal and vertical axis direction in the embodiment of the present invention.

2 is an exemplary view showing pixel positions of adjacent blocks in an embodiment of the present invention.

를 결정하기 위한

값을 보인 표. Figure 3 shows an embodiment of the present invention.

To determine

Table showing values.

4 is an operational flowchart for an embodiment of the present invention.

TECHNICAL FIELD The present invention relates to video compression techniques, and more particularly, to a method of filtering image pixels.

The H.26L method, which has been developed as the next generation video compression method since 1997, has proved superior in performance over the existing H.263 and MPEG4 methods.

H.26L, the next generation video compression method, uses 4 * 4 block-based transform and encoding, performs motion estimation and compensation of transform block size, and uses one variable length coder (VLC). This is a major difference from 263 and MPEG4.

As mentioned above, the H.26L method is superior to the existing video standardization method in terms of performance. However, since the encoder calculates a lot and the conversion method is based on the 4 * 4 block size, the blockage phenomenon is more than the conventional standardization method. This has a noticeable drawback.

The filter type for removing such a blocking phenomenon may be configured by a loop filter method which is processed inside the encoder and decoder, a post filter that is processed after the decoder, and a combination of the above two methods. .

In general, removing the blocking effect by the loop filter method affects the difference value between the input video and the motion video, and increases the coding rate in a specific case. However, since it is installed inside the video coder, the complexity of the coding unit is increased.

That is, the complexity of the loop filter is determined by the number of filter taps and the feasibility of the algorithm. Therefore, the loop filter has a small number of filter taps and a simple algorithm is preferable.

In addition, the post filter for removing the blocking phenomenon has the advantages and disadvantages as opposed to the loop filter.

However, despite the advantages and disadvantages of each filter method, the H.26L standardization mechanism currently considers only loop filters for standardization.

Accordingly, an object of the present invention is to provide a method for filtering an image pixel created to guarantee a certain level of image quality by removing a block and ring phenomenon of a compressed video as well as facilitating real-time processing by reducing a calculation amount in implementing a compressed video standardization method. There is this.

In particular, the present invention facilitates real-time processing by filtering pixels in the block boundary area based on at least one of motion vector information, transform coefficient information, and encoding type information, and removes blocking and ring phenomena to improve the quality of the compressed video. The purpose is to ensure.

In order to achieve the above object, the present invention provides a method of filtering pixels in a block boundary region of an image, comprising: a first process of obtaining motion vector information, transform coefficient information, and encoding type information of a block in contact with a block boundary; And to filter the pixels of the block boundary region based on a filtering method including a second process of determining whether to perform filtering on the block boundary region based on at least one of the information obtained in the first process.

The second process includes comparing motion vectors of two blocks adjacent to a block boundary to obtain a difference, determining whether transform coefficient information includes non-zero transform coefficient information, and encoding And determining whether the type is intra coding or inter coding.

The second process may include determining whether to perform filtering based on motion vector information, transform coefficient information, and encoding type information, and determining filtering strength based on the motion vector information, transform coefficient information, and encoding type information. Characterized in that it comprises at least one further step.
According to an aspect of the present invention, there is provided a method of filtering pixels in a block boundary region of an image to achieve the above object, the method comprising: quantizing at least one of motion vector information, transform coefficient information, and encoding type information associated with a block adjacent to the block boundary region; Determining whether to perform filtering on the block boundary region based on information; And performing filtering on the pixels of the block boundary area according to the determination result.
The encoding type information may be configured to include an intra encoding or an inter encoding type.
The motion vector information is a motion vector difference between two blocks adjacent to the block boundary area.
The transform coefficient information may include non-zero transform coefficient information.

The filtering may be performed in a horizontal or vertical direction with respect to the pixel.

That is, according to the present invention, in filtering pixels in the boundary region of an image block, whether to perform filtering or determining filtering intensity is performed based on at least one of motion vector information, transform coefficient information, and encoding type information. By performing filtering in the horizontal direction or the vertical direction of the image, the blocking phenomenon existing in the boundary area of the macro block is implemented.

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

The TML5 video compression scheme transmits temporally and spatially compressed information and additional information necessary for decoding by using a technique of removing spatial and temporal excess information from an encoder.

The decoder is configured to be executed in reverse order with respect to the operation of the encoder.

First, a real-time process of predicting an original image according to a quantization variable (QP), a coded block pattern, and a motion vector size that can use correlation between pixels to remove the block and ring phenomenon The technique will be explained.

When the original image f is transmitted in a compressed form, the image reconstructed by the decoding end may be expressed as follows.

---------------- (식1-1)

---------------- (Equation 1-1)

In Equation 1-1, g, f, and n denote column vectors of the compressed image, the original image, and the quantization error, which are rearranged in the scanning order.

(Equation 1-1) may be expressed as in Equation (1-2) below, which expresses the relationship of each pixel for pixel unit processing.

-------------- (식1-2)

-------------- (Equation 1-2)

When the reliability of the original pixel and the one-dimensional smoothing function for each pixel are used together with the normalization technique, an additional function such as Equation (1-3) can be defined. First consider the horizontal direction.

-----------(식1-3)

----------- (Equation 1-3)

,

은 화소 f(i,j)에 대한 좌측, 우측 방향으로의 신뢰도 평탄화 정도를 나타내는 함수이며, 이 함수들은 다음과 같이 정의된다.In the formula (1-3)

,

Is a function indicating the degree of reliability planarization in the left and right directions for the pixel f (i, j), and these functions are defined as follows.

-----(식1-4)

----- (Eq. 1-4)

에 대한 인접 화소와의 불규칙을 나타내고, 둘째 항들은 원 화소에 대한 신뢰도를 나타낸다. The first terms of the right term of each function of Equation (1-4) are pixels

Denotes an irregularity with adjacent pixels with respect to the second term, and the second terms indicate reliability with respect to the original pixel.

,

은 각 함수에 정의된 불규칙 항과 신뢰도 항의 비율을 나타내는 일종의 정규화 매개 변수들을 의미한다. Also,

,

Is a kind of normalization parameter representing the ratio of the irregular term and the confidence term defined in each function.

In this manner, an additional function is defined for each pixel of the video, and the function definition in the vertical direction is defined by changing i instead of the variable j in Equation 1-4.

Subsequently, in order to obtain a reconstructed image from which blocking and ringing are removed, the derivative of the above-described addition function defined in each pixel is obtained as follows.

---------------------(식1-5)

--------------------- (Equation 1-5)

In the above formula (1-5),

----(식1-6)

---- (Eq. 1-6)

Pixels f (i, j) to be restored in the horizontal direction from Equations 1-5 and 1-6 can be obtained as follows.

-----------(식1-7)

----------- (Equation 1-7)

Since the normalization parameter of Equation 1-7 has a value between 0 and 1, Equation 1-7 can be defined as follows.

-----------(식1-8)

----------- (Equation 1-8)

Referring to Equation 1-8, the deblocking image of the compressed image is determined by two pixel values and a normalization parameter based on the (i, j) th image.

Since the left and right pixel values are values that can be used by the encoder and the blocker, two normalization parameter values may be set to obtain a reconstructed image.

For this purpose, applying set theoretic theory to the pixel-based reconstruction method, normalization coefficients can be expressed as

-------------(식1-9)

------------- (Equation 1-9)

,

은 블록 경계 및 블록 내부의 화소간의 불균일 정도가 다르기 때문에 화소의 위치에 의해 결정되는 적응 매개 변수로서 통계적으로 결정되며, QP는 f(i,j)가 속한 매크로 블록의 양자화 매개 변수이다. Of the above formula (1-9)

,

Is statistically determined as an adaptive parameter determined by the position of the pixel because the degree of nonuniformity between the block boundary and the pixel inside the block is different, and QP is a quantization parameter of the macro block to which f (i, j) belongs.

Using Equation 1-8 and Equation 1-9, a reconstructed image obtained by removing the blocking phenomenon can be obtained, but since the value of the normalization parameter requires a floating-point operation, there is a problem in terms of computation. Can be.

Therefore, the above formulas (1-8) and (1-9) are modified as follows for integer arithmetic.

---- (식1-10)

---- (Eq. 1-10)

And,

----------(식1-11)

---------- (Eq. 1-11)

The algorithm of the present invention is used to generate a normalization parameter in the form of an integer using Equation 1-11 and then store it in the form of a lookup table.

Therefore, the above equation (1-10) for block filtering can be obtained by knowing the position of the pixel f (i, j) and the surrounding two pixels and the quantization parameter value.

Meanwhile, the loop filtering process of the present invention is performed in the horizontal axis direction to obtain an image from which the blocking phenomenon is removed in the horizontal axis direction, and to perform the image in the vertical axis direction to obtain the image from which the blocking phenomenon in the vertical axis direction is removed, respectively. The first step of determining the strength of the block to be processed according to the coding pattern and the quantization step size of each 4 * 4 block in the horizontal, vertical axis direction processing of Fig. 4, and the angle determined in the step A second step of performing a loop filtering process according to the strength of the block and a third step of removing a separate blocking phenomenon from the boundary area of the macro block are performed because a relatively large blocking phenomenon exists in the boundary area of the macro block.

Here, the positions of the blocks in the horizontal axis and vertical axis directions are the same as those of FIGS. 1A and 1B.

First, a first step of determining the block strength St and the filtering state will be described.

This is a process to determine the H.26L method as a standardized method. Currently, TML (Test Model Long Term) is in progress and the TML is compressed based on 4 * 4 blocks.

Of course, in order to remove the excess information in the time domain, the conventional MPEG and H.263 motion compensation schemes are used.

In order to remove the blocking phenomenon by using the basic mechanism of the TML method, it is generally processed in 4 * 4 block units.

In the present invention, the method for removing the blocking phenomenon follows the conventional method as described above, but the degree of the blocking phenomenon occurs differently according to the coding pattern, the quantization step size, and the motion vector size of each block. It is determined by the pretreatment process.

)를 '0'으로 초기화한다. (

)1. Strength of each 4 * 4 block (

) Is initialized to '0'. (

)

2. Each 4 * 4 block is an intra-coded block or has a non-zero transform coefficient

3. If the motion vector difference between two blocks block1 and block2 in the example of FIG. 1 is greater than or equal to '1'

는 a 및 b의 최대값을 의미한다. here,

Means the maximum value of a and b.

)를 결정한 후 각 블록의 양자화 크기에 따라 루프 필터링을 실행하게 된다. Also, as above, the strength of each block (

), Loop filtering is performed according to the quantization size of each block.

In the case of performing the loop filtering, the quantization size of two blocks block1 and block2 in the example of FIG. 1 is as follows.

1. The quantization value of the block block1 is smaller than '21' and the strengths St of the two blocks block1 and block2 are not both '0'.

2. The quantization value of the block block1 is greater than 20 and the intensity St of one of the two blocks block2 is not '0'.

The reason why the filtering state is determined differently according to the quantization size is to perform an adaptive process because the blocking phenomenon is prominent when the quantization size is large.

Therefore, when the condition of the preprocessing step is satisfied, the second step of filtering the loop to remove the blocking phenomenon is performed as follows.

When the pixel positions of two blocks block1 and block2 are as shown in the exemplary diagram of FIG. 2, the pixel c, d, e, and f values when the filtering state of each block defined above are satisfied are as follows. It is calculated by (2-1).

----------------------- 식(2-1)

----------------------- Formula (2-1)

는 다음과 같이 정의된다. In the formula (2-1)

Is defined as

------- 식(2-2)

------- Formula (2-2)

는 다음과 같이 정의된다. here,

Is defined as

---------------- 식(2-3)

---------------- Formula (2-3)

는 아래의 식(2-4)와 같이 정의되므로 도3의 표에 예시된

값을 대입하여 구할 수 있다. And,

Is defined as in Equation (2-4) below,

Can be found by assigning a value.

--------------- (2-4)

--------------- (2-4)

,

(또는

),

각각은 식(2-2)의

,

,

각각과 동일한 파라메터이다. By the way, said Formula (1-11) and Formula (2-2) are the same Formula, and

,

(or

),

Each of equation (2-2)

,

,

Same parameter as each.

Accordingly, the filtered values of the pixels c, d, e, and f are calculated as follows using the parameters defined above.

---------------- 식(2-5)

---------------- Formula (2-5)

은 비트 시프트(bit shift)를 의미한다. In the above formula (2-5)

Means bit shift.

However, as a result of the above process, the image satisfactorily removes the blocking and ringing phenomenon, but it can be confirmed that the blocking phenomenon still exists in the boundary region of the macroblock composed of 16 * 16 blocks. have.

In order to solve this problem, a third step of performing a separate filtering process such as the following Equation (2-6) is performed on the pixels located at the macroblock boundary.

------------------- 식(2-6)

------------------- Formula (2-6)

and (d,e) are macroblock boundary pixels.If

and (d, e) are macroblock boundary pixels.

As described in detail above, the present invention provides real-time processing by filtering pixels in a block boundary area based on at least one of motion vector information, transform coefficient information, and encoding type information in a digital video apparatus using H.26L video compression. In addition to making it easy, there is an effect of ensuring the quality of the compressed video by removing block and ring phenomenon.

In particular, the present invention has the effect of improving the image quality of a compressed image that requires low bit rate or high speed processing.

Claims (9)

In the method of filtering the pixels of the block boundary region of the image, Determining whether to perform filtering on the block boundary region based on at least one of motion vector information, transform coefficient information, and encoding type information related to a block adjacent to the block boundary region, and quantization information; And filtering the pixels of the block boundary area according to the determination result. The method of claim 1, wherein the encoding type information is An image pixel filtering method comprising an intra encoding or an inter encoding type. delete The method of claim 1, wherein the motion vector information is And a motion vector difference between two blocks adjacent to the block boundary area. The method of claim 1, wherein the transform coefficient information And non-zero transform coefficient information. delete delete delete delete

Images