KR101076272B1 - Method for extracting region of interest based on pixel domain for distributed video coding and computer readable medium thereof - Google Patents

Abstract

The present invention relates to distributed video coding and a technique for extracting a region of interest from an image.
According to the present invention, a method for extracting a region of interest (ROI) for distributed video coding (DVC) may include generating a mean absolute difference (MAD) map, and using SAD (MAD map). sum of absolute difference) generating a distribution chart and determining a region of interest using the SAD distribution chart.
According to the present invention, it is possible to extract a region of interest for distributed video coding, to reduce complexity of an encoder and a decoder, and to reduce a bit rate of processing.

Description

FIELD OF EXTRACTING REGION OF INTEREST BASED ON PIXEL DOMAIN FOR DISTRIBUTED VIDEO CODING AND COMPUTER READABLE MEDIUM THEREOF}

The present invention relates to distributed video coding and a technique for extracting a region of interest from an image.

MPEG and H.26x technologies, which are the current video standard compression techniques, use inter-frame predictive coding using continuous frame correlation to increase the compression rate. Due to the motion estimation used in this technique, the complexity of the encoder is generally about 5 to 10 times higher than the complexity of the decoder. This asymmetric video structure is well suited for broadcast systems or video on demand that compress once and decode multiple times.

However, recently demanded video applications, such as wireless surveillance video sensors, wireless PC cameras, portable terminal cameras, and disposable video cameras, have limited memory and power and have high computational limitations due to limitations in computing power. It is difficult to perform algorithms of complexity.

Therefore, new application products needed a new concept of video encoding technology that can encode video with low complexity, not predictive coding technology which is a conventional video coding method. Recently, distributed video coding (DVC) has attracted attention as a video coding technology suitable for new applications.

Over the years, distributed video coding has been studied based on two information theories published in the 1970s. One of them is the Slepian-Wolf theorem, which was released in 1973, and it is possible to achieve the same compression efficiency as the joint coding of two sources if the joint decoding is performed even if the correlated sources X and Y are independently encoded. Slepian-Wolf code means distributed source coding and is lossless compression. The other is the Wyner-Ziv theorem, published in 1976, which extends the Slepian-Wolf code to provide the theoretical boundary of lossy compression when using side information (SI) in the decoder.

When the image frames are independently encoded using this distributed source coding, the complexity of the encoder can be reduced by compressing without complicated processing such as calculating statistical dependencies between frames such as motion compensation and motion prediction. Can be lowered. In addition, current distributed video coding can achieve performance close to conventional intra coding such as MPEG.

In 2002, Wyner-Ziv (WZ) video, the actual design structure of distributed video coding, was released and distributed video coding has been extensively studied. The great advantage of WZ video coding is that the complexity of the encoder is reduced by compressing each frame independently of each other without using the inter-frame predictive coding technique that was complicatedly performed in the encoder.

However, the complexity of the decoder increased as the complexity of the encoder decreased. That is, a decoder is used to generate auxiliary information between frames, thereby increasing the complexity of the decoder.

To reduce the complexity of the decoder of distributed video coding, it is necessary to reduce the complexity of the decoding itself or solve the decoding delay problem caused by the feedback channel. As the latter method, there are an encoder control method for estimating and compressing an appropriate bit rate of a WZ frame at an encoder and a heterogeneous control method for bit rate estimation at an encoder and a decoder. However, these methods deal only with the decoding delay problem caused by feedback, and there is no algorithm which reduces the direct complexity of the encoder and decoder.

The present invention proposes an algorithm for reducing the complexity of low distributed video coding. We propose an algorithm that selectively extracts regions of interest (ROI) from WZ frames and selectively encodes them.

In the WZ frame, there is an area where the change of movement is small, such as the background of an image, and there is an area where the change of movement is large, such as an object or a person. Therefore, instead of encoding the entire WZ frame, an encoding region is selectively selected for the region of interest by defining a region of interest in which a lot of motion or a change of movement is severe in the WZ frame. This reduces the bit rate of the WZ frame, reduces the rate distortion, and reduces the complexity of the encoder and decoder.

An object of the present invention is to extract a region of interest for distributed video coding, to reduce complexity of an encoder and a decoder, and to reduce a bit rate of processing.

A method of extracting a region of interest (ROI) of an image according to the present invention includes generating a mean absolute difference (MAD) map of an image, and generating a SAD (sum of absolute difference) distribution map using the MAD map. And determining a region of interest using the SAD distribution chart.

MAD map is generated by using the average value of absolute difference between previous key frame and next key frame, and each pixel included in MAD map and key frame has height position information and width position information. It is preferable to include.

The SAD distribution chart includes a vertical SAD distribution chart and a horizontal SAD distribution chart. A vertical SAD distribution chart is generated by adding MAD values of each pixel included in the MAD map at a predetermined width position, and the horizontal SAD distribution chart is a MAD map at a predetermined height position. It is preferable to generate the MAD value included in each pixel included in the horizontal direction.

The ROI is a rectangle having a height smaller than the height of the entire image and a width smaller than the width of the entire image. Each pixel included in the ROI includes height position information and width position information, and height position information of the ROI. Indicates the height position of the ROI where the sum of the horizontal SAD values belonging to the height of the ROI is the largest, and the width position information of the ROI is the position where the sum of the vertical SAD values belonging to the width of the ROI is the largest. It is preferable to indicate the width position of the region of interest.

The height location information of the ROI includes height location information of the top pixel of the ROI where the sum of the horizontal SAD values is the largest, and the width location information of the ROI is the location where the sum of the vertical SAD values is the largest. It is preferable to include height position information of the leftmost pixel of the ROI.

A method of extracting a region of interest (ROI) for an image according to the present invention includes generating an absolute residual map (ARM), generating a sum of absolute difference (SAD) distribution using the ARM, and using the SAD distribution map. Determining a step.

The ARM is generated using the absolute difference between the previous keyframe and the next keyframe, and each pixel included in the ARM and the keyframe preferably includes height position information and width position information.

The SAD distribution chart includes a vertical SAD distribution chart and a horizontal SAD distribution chart. A vertical SAD distribution chart is generated by adding ARM values of each pixel included in the ARM at a predetermined width position, and the horizontal SAD distribution chart is included in the ARM at a predetermined height position. It is preferable to generate by adding the ARM value of each pixel.

The ROI is a rectangle having a height smaller than the height of the entire image and a width smaller than the width of the entire image. Each pixel included in the ROI includes height position information and width position information, and height position information of the ROI. Indicates the height position of the ROI where the sum of the horizontal SAD values belonging to the height of the ROI is the largest, and the width position information of the ROI is the position where the sum of the vertical SAD values belonging to the width of the ROI is the largest. It is preferable to indicate the width position of the region of interest.

The height location information of the ROI includes height location information of the top pixel of the ROI where the sum of the horizontal SAD values is the largest, and the width location information of the ROI is the location where the sum of the vertical SAD values is the largest. It is preferable to include height position information of the leftmost pixel of the ROI.

A method of extracting a region of interest (ROI) for an image according to the present invention includes dividing an image into M blocks (where M is an integer), calculating a block SAD (sum of absolute difference) of each block, and a block SAD value. Selecting a region of interest block using;

The block SAD obtains the difference between the previous key frame and the next key frame by adding the height and width of the block, and preferably, each pixel included in the block SAD and the key frame includes the height position information and the width position information. .

The region of interest block is preferably selected from N (1 <N ≤ M, N is an integer) in order of increasing block SAD value.

A computer readable medium including computer executable instructions for causing a computer to perform the following steps, extracting a region of interest (ROI) of an image, comprises: generating a MAD map, using a SAD map Generating a distribution chart and determining a region of interest using the SAD distribution map.

A computer readable medium including computer executable instructions for causing a computer to perform the following steps, wherein the computer readable medium for extracting a region of interest (ROI) for an image comprises: generating an ARM, a SAD distribution diagram using the ARM And generating a region of interest using the SAD distribution map.

A computer readable medium including computer executable instructions for causing a computer to perform the following steps, wherein the computer readable medium for extracting a region of interest (ROI) for an image comprises M blocks (where M is an integer). Dividing, calculating a block SAD of each block, and selecting a region of interest block using a block SAD value, wherein the block SAD takes an absolute value of a difference between a previous key frame and a next key frame. Each pixel included in the block SAD and the key frame includes height position information and width position information.

According to the present invention, it is possible to extract a region of interest for distributed video coding, to reduce complexity of an encoder and a decoder, and to reduce a bit rate of processing.

1 is a diagram illustrating a region of interest extraction method in distributed video coding according to an embodiment of the present invention.
2A is a diagram illustrating an image 200 used in a method for extracting an entire region of interest unit of interest according to an embodiment of the present invention.
2B is a flowchart illustrating a method of extracting an entire region of interest unit of interest according to an embodiment of the present invention.
3A is a diagram illustrating a vertical SAD distribution diagram of an image 200 used in an ROI extraction method of an entire image unit according to an embodiment of the present invention.
3B is a diagram illustrating a horizontal SAD distribution diagram of an image 200 used in an ROI extraction method of an entire image unit according to an exemplary embodiment.
FIG. 4 is a diagram illustrating a result of extracting a region of interest with respect to a hall monitor image by using a method of extracting a region of interest in an entire image unit according to an embodiment of the present invention.
5 is a flowchart illustrating a method of extracting a region of interest region per frame according to an embodiment of the present invention.
FIG. 6A is a diagram illustrating an image using a method of extracting a block of interest region according to an embodiment of the present invention. FIG.
6B is a flowchart illustrating a method of extracting an ROI in a block unit according to an embodiment of the present invention.
FIG. 6C is a diagram illustrating a result of extracting a region of interest using a block-based region of interest extraction method according to an embodiment of the present invention. FIG.

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, detailed descriptions of preferred embodiments of the present invention will be described with reference to the accompanying drawings. It should be noted that reference numerals and like elements among the drawings are denoted by the same reference numerals as much as possible even though they are shown in different drawings.

In describing the present invention, when it is determined that the detailed description of the related known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted.

Distributed Video Coding Using Region Extraction distributed video coding , DVC )

1 is a diagram illustrating a region of interest extraction method in distributed video coding according to an embodiment of the present invention.

)과 Wyner-Ziv(WZ) 프레임(

)으로 나누어 처리한다. 여기서 홀수 프레임과 짝수 프레임의 구별은 프레임 카운팅을 시작하는 시간 색인을 0으로 할 것인지, 1로 할 것인지에 따라 구별되며, 이하 본 발명의 명세서에서는 프레임 카운팅을 시작하는 색인을 1로 하여 설명할 것이며, 따라서 키프레임의 시간 색인은 홀수가 되고, WZ 프레임의 시간 색인은 짝수가 된다. 키프레임의 시간 색인이 짝수가 되고, WZ 프레임의 시간 색인이 홀수가 될 수도 있다.키프레임과 WZ 프레임은 서로 독립적으로 부호화되며, H.264 인트라 인코더(intra encoder, 101)는 키프레임을 H.264 인트라 부호화를 한다.In the encoding process of pixel domain distributed video coding, as in the conventional distributed video coding, a frame image is classified into odd frames and even frames, and each of them is keyframe (

) And Wyner-Ziv (WZ) frames (

Divided by) Here, the distinction between the odd frame and the even frame is distinguished according to whether the time index for starting frame counting is 0 or 1, and in the following specification, the index for starting frame counting will be described as 1, Therefore, the temporal index of the keyframe becomes odd, and the temporal index of the WZ frame becomes even. The temporal indexes of the keyframes may be even, and the temporal indexes of the WZ frames may be odd. The keyframes and the WZ frames are encoded independently of each other, and the H.264 intra encoder 101 encodes the keyframes in H. Perform .264 intra coding.

)을 추출하여, 추출된 관심영역(

)을 부호화부(103)에 전달한다. 관심영역(

)을 추출하는 구체적인 방법은 아래에서 설명한다.The ROI extractor 102 may encode an ROI in order to encode a WZ frame.

), The extracted ROI (

) Is transmitted to the encoder 103. Area of interest (

The specific method of extracting a) is described below.

)만을 부호화하여 압축한다.The encoder 103 extracts the extracted ROI (

) Is encoded and compressed.

,

)을 보조정보 생성부(105)로 전달한다.In the decoding process of the pixel domain distributed video coding, the H.264 intra decoder 104 performs H.264 intra decoding, and the decoded keyframe (

,

) Is transmitted to the auxiliary information generator 105.

,

)을 이용하여 WZ 프레임과 닮은 보조정보(

)를 생성하게 된다.The auxiliary information generating unit 105 is a decoded key frame (

,

), Which is similar to the WZ frame,

Will be generated.

) 및 부호화부(103)에서 얻어진 WZ 프레임의 압축된 정보(

)를 이용하여 WZ 프레임의 비트 스트림(

)을 생성 된다.The decoding unit 106 generates the auxiliary information generated by the auxiliary information generating unit 105 (

) And the compressed information of the WZ frame obtained by the encoder 103 (

), The bit stream of the WZ frame (

Is generated.

) 및 보조정보(

)를 이용하여 최종적으로 WZ 프레임(

)을 복원한다.
As a last step, the reconstruction unit 107 performs a bit stream (

) And ancillary information (

) And finally the WZ frame (

Restore).

Whole video unit of interest region of interest , ROI ) Extraction Method

2A is a diagram illustrating an image 200 used in a method for extracting an entire region of interest unit of interest according to an embodiment of the present invention.

Image 200 is a hall monitor image, which is two people 201 and 202 located in the center area of the corridor, walls 205 and 206 on both sides of the corridor, and three doors 203, 204 and 207. ).

In general, when the ROI is extracted from the hall monitor image, it is preferable to extract the center area of the corridor with a lot of motion as the ROI. However, this is an intuitive result and the following describes how to find a region of interest in a systematic manner.

2B is a flowchart illustrating a method of extracting an entire region of interest unit of interest according to an embodiment of the present invention.

The entire image unit ROI extraction method refers to a method of extracting an ROI for the entire image 200 before encoding a Wyner-Ziv (WZ) frame. The method of extracting the whole image unit of interest region is performed by the following procedure.

First, a MAD (mean absolute difference) map is generated between frames of the entire image (S201).

Thereafter, a SAD distribution is generated using the generated MAD map (S202).

After that, the region of interest is determined using the generated SAD distribution (S203).

Hereinafter, each step will be described in detail.

MAD ( mean absolute difference Create map

Mean absolute difference (MAD) means the average value of absolute differences.

Each pixel in the MAD map and keyframe contains height and width location information, and the MAD map is the absolute difference between the previous keyframe and the next keyframe of the entire image. Calculate using the average value of.

As shown in Equation (1) below, the difference between the absolute values of the key frames is obtained and the sum is added to find the average.

...(1)

...(One)

은 시간 색인이 2n-1 번째인 키프레임을 의미하고,

는 2n-1 번째 프레임의 높이위치정보인 x, 너비위치정보인 y인

위치의 픽셀 값을 의미한다. 그리고 N은 전체 키프레임의 개수를 의미하고,

는 각각 프레임의 높이(height)와 너비(width)를 의미한다.here

Means a keyframe with a time index of 2n-1th,

X is the height position information of the 2n-1th frame and y is the width position information

The pixel value of the position. And N is the total number of keyframes,

Are the height and width of the frame, respectively.

을 높이위치정보(x), 너비위치정보인 (y)로 2차원 맵(map)으로 나타낸 것이 MAD 맵이다.
Obtained by Equation (1)

MAD map is represented by height position information (x) and width position information (y) as two-dimensional maps.

SAD  Generate distribution

The SAD (sum of absolute difference) distribution shows the distribution of the sum of the absolute values according to the position of a certain standard.

를 이용하여 구하며, SAD 분포도는 수직 SAD 분포도와 수평 SAD 분포도가 있다.The SAD distribution map was obtained from the process of generating the MAD map.

The SAD distribution chart has a vertical SAD distribution and a horizontal SAD distribution.

3A is a diagram illustrating a vertical SAD distribution diagram of an image 200 used in an ROI extraction method of an entire image unit according to an embodiment of the present invention.

에서 일정기준을 너비(y)로 두고, 특정한 너비(y)에 대응하는 높이(x)값을 수직으로 더한다.The vertical SAD distribution map is generated by adding the height position information of each pixel included in the MAD map in the vertical direction. In other words,

In Figure 3, a certain criterion is defined as width (y), and the height (x) corresponding to a specific width (y) is added vertically.

The specific formula is shown in the following formula (2.a).

...(2.a)

... (2.a)

Looking at the vertical SAD distribution, the width (y) of the SAD value 301 of about 30 to 110 parts is relatively larger than the surroundings. As a result, the width (y) of 30 to 110 of the whole image will be the width position of the ROI. The determination of the location of the ROI is described below by finding the ROI in a systematic manner.

3B is a diagram illustrating a horizontal SAD distribution diagram of an image 200 used in an ROI extraction method of an entire image unit according to an exemplary embodiment.

에서 일정기준을 높이(x)로 두고, 특정한 높이(x)에 대응하는 너비(y)값을 수평으로 더한다.The horizontal SAD distribution map is created by adding the width position information of each pixel in the horizontal direction. In other words,

Set a certain criterion as height (x) in, and horizontally add the width (y) value corresponding to the specific height (x).

The specific formula is shown in the following formula (2.b).

...(2.b)

... (2.b)

Looking at the horizontal SAD distribution, the height (x) of approximately 30 to 120 parts of the SAD value 302 is relatively larger than the surroundings. Eventually, the height (height, x) of 30 to 120 in the whole image will be the height position of the ROI. The determination of the height location of the ROI is described below by finding the ROI in a systematic manner.

Area of interest ( region of interest , ROI ) decision

와 수평 SAD 분포도

를 기반으로 관심영역의 위치

를 결정하는 것이다.ROI determination is based on vertical SAD distribution

And horizontal SAD distribution

Location of region of interest

To determine.

)를 찾기 위해서는 아래 수식 (3.a)와 같이 먼저 도 3b에 나타낸 수평 SAD 분포도에서 관심영역의 높이 (

) 크기의 창을 처음부터 끝까지 이동하면서 그 창에 속하는 SAD 값들의 합이 가장 큰 창을 찾아야 한다. 그 후, SAD 값들의 합이 가장 큰 창의 첫 높이가 바로 관심 영역의 세로 위치(

)이다.The height of the region of interest

) To find the height of the region of interest in the horizontal SAD distribution plot shown in Figure 3b, as shown in equation (3.a) below:

As you move the window of size from beginning to end, you must find the window with the largest sum of SAD values that belong to that window. The first height of the window with the largest sum of SAD values is then the vertical position of the region of interest (

)to be.

...(3.a)

... (3.a)

)를 찾기 위해서는 아래 수식 (3.b)와 같이 도 3a에 나타낸 수직 SAD 분포도에서 관심영역의 너비 (

) 크기의 창을 처음부터 끝까지 이동하면서 그 창에 속하는 SAD 값들의 합이 가장 큰 창을 찾아야 한다. 그 후, SAD 값들의 합이 가장 큰 창의 첫 너비 위치가 바로 관심 영역의 너비 위치(

)이다.Also, in the same way, the width position of the region of interest (

), The width of the region of interest in the vertical SAD distribution plot shown in Figure 3a as shown in equation (3.b) below.

As you move the window of size from beginning to end, you must find the window with the largest sum of SAD values that belong to that window. The first width position of the window with the largest sum of SAD values is then the width position of the region of interest (

)to be.

...(3.b)

... (3.b)

와

는 각각 관심영역의 높이크기와 너비크기를 의미한다.here

Wow

Are the height and width of the ROI, respectively.

FIG. 4 is a diagram illustrating a result of extracting a region of interest with respect to a hall monitor image by using a method of extracting a region of interest in an entire image unit according to an embodiment of the present invention.

When the region of interest is extracted from the hall monitor image from the whole image 200, it is preferable to extract the center region of the corridor with a lot of motion as the region of interest. Likewise, the region of interest 410 is determined.

) 및 너비크기(

)는 임의로 조정이 가능한 값이고, 일반적으로는 전체영상의 높이크기 및 너비크기(

)의 절반으로 설정한다.Where the height of the region of interest (

) And width size (

) Is a value that can be arbitrarily adjusted, and in general, the height size and width size (

Set to half of).

According to the present invention, since the key frame is information that the encoder and the decoder have at the same time, the ROI can be extracted using only the key frame without any other information. Therefore, even if the encoder does not inform the decoder about the region of interest, the decoder can extract the region of interest independently from the encoder and use it in the decoding process.

frame( frame Unit of interest ( region of interest , ROI ) Extraction Method

The above-described method of extracting a region of interest of an entire image unit is performed only once before encoding the entire image, but the method of extracting a region of interest of a frame unit is performed each time before encoding each frame. The method of extracting an ROI in a frame unit is performed by the following procedure.

5 is a flowchart illustrating a method of extracting a region of interest region per frame according to an embodiment of the present invention.

First, an absolute residual map (ARM) is generated for each frame (S501).

Then, using the generated ARM to generate a sum of absolute difference (SAD) distribution (S502).

After that, the region of interest is determined using the generated SAD distribution (S503).

Hereinafter, each step will be described in detail.

ARM ( absolute residual map ) produce

번째 프레임을 부호화 한다고 가정하면, 아래 수식 (4)와 같이 이전(

) 프레임과 다음(

) 프레임의 절대값 차이인 ARM(Absolute Residual Map)을 생성한다.first

Assuming that the first frame is to be encoded, the previous (

) Frame and the next (

Absolute Residual Map (ARM), which is the absolute difference of frames, is generated.

...(4)

...(4)

는 시간 색인인 n 번째 프레임을 의미하고,

은 n번째 프레임의 높이위치정보(x) 및 너비위치정보(y)의 위치

의 픽셀값을 의미한다.
here

Means the nth frame, which is the time index,

Is the position of height position information (x) and width position information (y) of the nth frame.

Means the pixel value of.

SAD  Generate distribution

The SAD (sum of absolute difference) distribution shows the distribution of the sum of the absolute values according to the position of a certain standard.

를 이용하여 구하며, SAD 분포도는 수직 SAD 분포도와 수평 SAD 분포도가 있다.SAD distribution was obtained from the previous ARM generation process.

The SAD distribution chart has a vertical SAD distribution and a horizontal SAD distribution.

The vertical SAD distribution map is included in ARM. It is generated by adding height position information of each pixel in the vertical direction. The specific formula is shown in the following formula (5.a).

...(5.a)

... (5.a)

The horizontal SAD distribution map is included in ARM. It is created by adding the width position information of each pixel in the horizontal direction. The specific formula is shown in the following formula (5.b).

...(5.b)

... (5.b)

Area of interest ( region of interest , ROI ) decision

와 수평 SAD 분포도

를 기반으로 관심영역의 위치

를 결정하는 것이다.Region of interest determination is a vertical SAD distribution

And horizontal SAD distribution

Location of region of interest

To determine.

)를 찾기 위해서는 아래 수식 (6.a)와 같이 먼저 수평 SAD 분포도에서 관심영역의 높이 크기(

)의 창을 처음부터 끝까지 이동하면서 창에 속하는 SAD 값들의 합이 가장 큰 창을 찾아야 한다. 그 후, SAD 값들의 합이 가장 큰 창의 높이 위치가 바로 관심영역 의 높이(

)이다.The height of the region of interest

) To find the height size of the ROI in the horizontal SAD distribution plot (6.a)

As you move through the windows of), you must find the window with the largest sum of the SAD values that belong to the window. Then, the height position of the window with the largest sum of SAD values is the height of the region of interest (

)to be.

...(6.a)

... (6.a)

)를 찾기 위해서는 아래 수식 (6.b)와 같이 먼저 수직 SAD 분포도에서 관심영역의 너비크기(

)의 창을 처음부터 끝까지 이동하면서 그 창에 속하는 SAD 값들의 합이 가장 큰 창을 찾아야 한다. 그 후, SAD 값들의 합이 가장 큰 창의 너비 위치가 바로 관심영역의 너비위치(

)를 찾는다.Also, in the same way, the width position of the region of interest (

) To find the width size of the region of interest in the vertical SAD distribution plot (6.b).

Move the window of) from start to finish and find the window with the largest sum of SAD values that belong to that window. After that, the width of the window with the largest sum of SAD values is the width of the region of interest (

Find).

...(6.b)

... (6.b)

block( block Unit of interest ( ROI ) Extraction Method

FIG. 6A is a diagram illustrating an image using a method of extracting a block of interest region according to an embodiment of the present invention. FIG.

There is a person 601 in the image and the area of the person with motion compared to the surrounding background is closer to the triangle 602 than to the rectangle.

The shape of the ROI of the entire image unit ROI extraction method and the frame unit ROI extraction method is fixed as a rectangle, and only the height and width of the rectangle can be changed. However, this is not suitable for an image as shown in FIG. 5, unless a region having a large change in motion is concentrated in a specific region in the whole image. That is, the entire image unit ROI extraction method and the frame unit ROI extraction method for extracting one large ROI are not preferable in terms of efficiency in the image of FIG. 5.

Therefore, in order to extract a region of interest suitable for various images in which a large change in motion is required, several small regions of interest must be extracted or the size of the region of interest must also be changed.

In order to solve the above problem, the frame image is divided into a plurality of blocks of small size, a block having a lot of motion is selected, and the selected blocks constitute a proposed block unit ROI.

In this case, the number of selected blocks is determined such that the total area of the selected blocks is equal to the size of the ROI extracted by using the above-described method of extracting the ROI and the image unit. The block-based ROI extraction method is performed by the following procedure.

6B is a flowchart illustrating a method of extracting an ROI in a block unit according to an embodiment of the present invention.

First, the whole image is divided into M blocks (M is an integer) (S601). Here, the number M of blocks can be arbitrarily determined, and the larger the number of blocks, the smaller the size of each block.

Thereafter, a block sum of absolute difference (SAD) of each block is calculated (S602).

Thereafter, the ROI blocks are selected using the calculated block SAD value (S603).

Hereinafter, each step will be described in detail.

block SAD ( sum of absolute difference ) Calculation

The block unit of interest region extraction method is performed every time before each frame is encoded, as in the frame unit region of interest extraction method described above.

번째 프레임을 부호화한다고 가정하면, 아래 수식 (7)과 같이 이전(

) 프레임과 다음(

) 프레임의 차이값의 절대값을 취하고 이를 블록의 높이 및 너비로 더하여 계산한다.first

Assuming that the first frame is to be encoded, the previous (

) Frame and the next (

) Is calculated by taking the absolute value of the difference of the frames and adding it to the height and width of the block.

...(7)

... (7)

은 시간 색인이 2n 번째 프레임을 의미하고, i는 각각의 분할된 블록의 높이위치를 의미하며, j는 각각의 분할된 블록의 너비위치를 나타낸다. i 및 j 값의 범위는

이 된다.here

Denotes a 2n-th frame of time index, i denotes a height position of each divided block, and j denotes a width position of each divided block. i and j values range

Becomes

Here, h means the size of the height of the divided block, w is the size of the width of the divided block.

Area of interest ( region of interest , ROI Block selection

In this step, the ROI block is selected using the block SAD value. Since the region with much motion is the region of interest, blocks having a large block SAD value of each block are selected. N blocks are selected in order of block SAD values. Where N is an integer, and the range is (1 <N ≦ M), where M is the total number of blocks for dividing the entire image.

In general, the number of blocks selected in order of increasing block SAD value is determined so that the total area of the selected blocks is equal to the size of the ROI extracted by using the above-described method of extracting ROIs by the entire image unit and frame.

FIG. 6C is a diagram illustrating a result of extracting a region of interest using a block-based region of interest extraction method according to an embodiment of the present invention. FIG.

The region of interest is extracted by using the plurality of small blocks 610 for the movement of the person 601, which is preferable in terms of efficiency than the method of extracting the region of interest by the whole image unit and the frame unit.

As described above, those skilled in the art to which the present invention pertains will understand that the present invention may be implemented in other specific forms without changing the technical spirit or essential features.

Therefore, the embodiments described above are to be understood in all respects as illustrative and not restrictive, and the scope of the present invention is indicated by the following claims rather than the above description, and the meaning and scope of the claims and All changes or modifications derived from the equivalent concept should be interpreted as being included in the scope of the present invention.

101: intra encoder
102: region of interest extraction unit
103: encoder
104: intra decoder unit
105: auxiliary information generator
106: decryption unit
401: region of interest
610: region of interest block

Claims (16)

As a method of extracting a region of interest (ROI) of an image,
Generating a mean absolute difference (MAD) map for the image;
Generating a sum of absolute difference (SAD) distribution map using the MAD map; And
And determining a region of interest using the SAD distribution map.
The method of claim 1,
The MAD map is generated using an average value of an absolute difference between a previous key frame and a next key frame.
Each pixel included in the MAD map and the key frame includes height location information and width location information.
The method of claim 2,
The SAD distribution chart includes a vertical SAD distribution chart and a horizontal SAD distribution chart,
The vertical SAD distribution map is generated by adding MAD values of each pixel included in the MAD map at a predetermined width position,
The horizontal SAD distribution map is generated by adding a MAD value of each pixel included in the MAD map in a horizontal direction at a predetermined height position.
The method of claim 3,
The ROI is a rectangle having a height smaller than the height of the entire image and having a width smaller than the width of the entire image.
Each pixel included in the ROI includes height location information and width location information.
The height position information of the ROI indicates a height position of the ROI in which the sum of the horizontal SAD values belonging to the height of the ROI is the largest.
The width position information of the region of interest indicates a width position of the region of interest in which the sum of the vertical SAD values belonging to the width of the region of interest is the largest position.
The method of claim 4, wherein
The height position information of the ROI includes height position information of the top pixel of the ROI where the sum of the horizontal SAD values is the largest.
The width position information of the ROI includes height position information of the leftmost pixel of the ROI where the sum of the vertical SAD values is the largest.
In the ROI extraction method for an image,
Generating an absolute residual map (ARM);
Generating a sum of absolute difference (SAD) distribution map using the ARM; And
And determining a region of interest using the SAD distribution map.
The method of claim 6,
The ARM is generated using the absolute difference between the previous keyframe and the next keyframe,
Each pixel included in the ARM and the key frame includes height position information and width position information.
The method of claim 7, wherein
The SAD distribution chart includes a vertical SAD distribution chart and a horizontal SAD distribution chart,
The vertical SAD distribution chart is generated by adding an ARM value of each pixel included in the ARM at a predetermined width position,
The horizontal SAD distribution chart is generated by adding an ARM value of each pixel included in the ARM at a predetermined height position.
The method of claim 8,
The ROI is a rectangle having a height smaller than the height of the entire image and having a width smaller than the width of the entire image.
Each pixel included in the ROI includes height location information and width location information.
The height position information of the ROI indicates a height position of the ROI in which the sum of the horizontal SAD values belonging to the height of the ROI is the largest.
Width position information of the region of interest indicates a width position of the region of interest in which the sum of the vertical SAD values belonging to the width of the region of interest is the largest;
10. The method of claim 9,
The height position information of the ROI includes height position information of the top pixel of the ROI where the sum of the horizontal SAD values is the largest.
The width position information of the ROI includes height position information of the leftmost pixel of the ROI where the sum of the vertical SAD values is the largest.
In the ROI extraction method for an image,
Dividing the image into M blocks (M is an integer);
Calculating a block sum of absolute difference (SAD) of each block; And
Selecting the ROI block using the block SAD value;
The block SAD is obtained by adding the difference between the previous keyframe and the next keyframe to the height and width of the block,
Each pixel included in the block SAD and the key frame includes height position information and width position information.
delete The method of claim 11,
And the region of interest block selects N pieces (1 <N ≦ M, N is an integer) in order of increasing block SAD value.
Claim 14 was abandoned when the registration fee was paid. A computer readable medium containing computer executable instructions for causing a computer to perform the following steps, wherein the computer readable medium extracts a region of interest (ROI) of an image, wherein the following steps comprise:
Generating a mean absolute difference (MAD) map;
Generating a sum of absolute difference (SAD) distribution map using the MAD map; And
And determining a region of interest using the SAD distribution map.
Claim 15 was abandoned upon payment of a registration fee. A computer readable medium comprising computer executable instructions for causing a computer to perform the following steps, wherein the computer readable medium extracts a region of interest (ROI) for an image, the following steps:
Generating an absolute residual map (ARM);
Generating a sum of absolute difference (SAD) distribution map using the ARM; And
And determining a region of interest using the SAD distribution map.
Claim 16 was abandoned upon payment of a setup registration fee. A computer readable medium comprising computer executable instructions for causing a computer to perform the following steps, wherein the computer readable medium extracts a region of interest (ROI) for an image, the following steps:
Dividing the image into M blocks (M is an integer);
Calculating a block sum of absolute difference (SAD) of each block; And
Selecting the ROI block using the block SAD value;
The block SAD is obtained by adding the difference between the previous keyframe and the next keyframe to the height and width of the block,
And each pixel included in the block SAD and a keyframe includes height position information and width position information.

Images