KR101535196B1 - Apparatus and Method of Image Segmentation based on the Texture Characteristics - Google Patents

Abstract

The present invention relates to a technique for segmenting an image, and more particularly, to an apparatus for segmenting an image based on text characteristics and a method thereof. According to the present invention, a texture channel signal which presents complexity of a texture based on a roughness level of an image is calculated by using a fractal-dimensional image signal, and the image is segmented based on a mean shift vector calculated by using the calculated texture channel signal such that a flat pattern area of the texture and a boundary of the image are stably detected and separated, and also the occurrence of over-segmentation for a part determined as the same texture with respect to even a complex pattern area is suppressed, and the part is effectively separated as one area.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image segmentation apparatus and an image segmentation method,

The present invention relates to a technique for segmenting an image.

Image segmentation or image segmentation is a technique that divides an image into several meaningful regions in the field of computer vision. For this purpose, the images are classified according to certain characteristics, and the regions having the same characteristics are combined into one, and the boundaries are set between the regions having different characteristics. Such a segmentation technique is used for dividing boundaries in an image and detecting objects using divided regions.

In order to achieve the above-mentioned object, there is a watershed method as a conventional technique proposed. The above method is widely used as a method of dividing a region of an image by setting a boundary where the distribution of the pixels in the image is lower or higher than the other region as an image boundary and adjusting the threshold value. However, the above method has a problem in that over-segmentation in the image, that is, excessive region division is performed in a portion where the texture is complicated.

For example, there is a method of performing image area segmentation in a manner that merges regions based on an oversegmented image by a Watershed algorithm as in the prior art document. However, the above method has a fundamental defect in the algorithm that performs the merging using the brightness and color information based on the oversegmentation image obtained using only the brightness information, and the problem that the area division is different according to the image merging threshold value have.

International Journal: Automatic Image Segmentation by Dynamic Region Merging, Bo Peng, Zhang, D., Image Processing, IEEE Transactions on (Volume: 20, Issue: 12)

SUMMARY OF THE INVENTION The object of the present invention is to provide a method and an apparatus for processing a texture channel signal representing a texture complexity based on roughness of an image using a fractal image signal and dividing the image based on an average motion vector calculated using the texture channel signal, It is possible to stably detect and distinguish the boundary between the flat area of the texture and the image, and to divide the area, which is determined to be the same texture, into a single area while suppressing the occurrence of oversegmentation, And an object thereof is to provide an image dividing device and a method thereof which take texture characteristics into account.

According to an aspect of the present invention, there is provided an image dividing apparatus that takes texture characteristics according to one aspect of the present invention into a plurality of subpixels, A texture channel signal acquisition unit for acquiring a texture channel signal representing a signal distribution characteristic of a regional image composed of pixels; An average shifting process (Mean Shift Process) is performed for each pixel of the target image using the coordinate information of each pixel, the video signal information, and the obtained texture channel signal to obtain an average vector Process part; And an image divider that divides the target image into partial images using the average vector obtained by the average movement process unit.

The image processing apparatus may further include a normalization unit performing a normalization process of adjusting a dynamic range of the texture channel signal obtained by the texture channel signal acquisition unit according to a dynamic range of a video signal of the target image, And the average movement process unit performs an average movement process using the texture channel signal normalized by the normalization unit.

Here, the texture channel signal obtaining unit may calculate a variance of pixels located within a predetermined distance from each pixel of each pixel of the target image, and set the calculated variance as the texture channel signal of each pixel . ≪ / RTI >

Here, the texture channel signal obtaining unit may perform a local operation on each pixel of the target image to obtain a fractal image, and set the signal value of the fractal image to the texture channel signal value .

Here, the texture channel signal obtaining unit obtains a luminance image of the target image, sets an image block having a predetermined size for each pixel of the luminance image, calculates a fractal size for the set image block, And calculating a signal value for each pixel of the fractal image using the calculated fractal dimension.

Here, the texture channel signal obtaining unit may normalize the calculated fractal dimension to a predetermined range, and calculate a signal value for each pixel of the fractal dimension image.

Here, the fractal dimension may be a box counting dimension.

Here, the average moving process unit obtains, for each pixel of the target image, a first vector component having a pixel coordinate value, a second vector component composed of a luminance signal, a color difference signal, and the texture channel signal, Peripheral pixels within a threshold distance from each of the pixels for each pixel of the target image and for which the difference between the second vector component with the pixel is within the threshold signal difference is selected and set as an average moving participant pixel for each pixel , And calculating the average vector of each pixel using the average moving interest pixel.

Here, the average movement process unit may use the first vector component and the second vector component of each pixel of the target image and the first vector component and the second vector component of the average motion-involved pixels for each pixel, , And the average vector is calculated by calculating the average vector of each pixel based on the first vector component composed of the pixel coordinate values, the luminance signal, the color difference signal, and the second vector component And a control unit.

Here, the average moving process unit may calculate the average vector through an iterative updating process.

Here, the initial updating process of the updating process sets an initial value of the average vector to a vector composed of the first vector component and the second vector component of each pixel, Calculating the new average vector using the first vector component and the second vector component of the pixel to update the average vector, and updating the initial vector after the first updating process, And updating the average vector using an average vector and the first vector component and the second vector component of the average moving-involved pixel.

Here, the average moving process unit may calculate a first distance between the first vector component of the average vector and the first vector component of the average moving-participating pixel, and calculate a second distance between the second vector component of the average vector and the average moving Calculating a second distance between the second vector components of the involved pixels, and updating the average vector using the first distance and the second distance.

Here, the average moving process unit may calculate the average vector according to the following equation 1, when a vector including the first vector component and the second vector component in a pixel is a specific vector: Image segmentation device considering characteristics.

Equation 1

는 상기 평균이동 관여화소의 상기 특정 벡터이고,

는 상기 문턱신호차이이고,

는 상기 문턱거리이고,

는 상기

의 상기 제1 벡터 성분이고,

는 상기

의 상기 제2 벡터 성분이고,

는 상기 X에 대하여 산출된 상기 평균벡터 V(X)의 제1벡터 성분이고,

는 상기 X에 대하여 산출된 상기 평균벡터 V(X)의 제2벡터 성분이다.)Wherein X is the specific vector in the target pixel from which the average vector is calculated, V (X) is the average vector, j is an index indicating the number of iterations of the updating process, and N (X) And a set of each of the specific vectors of the average mobile-affixed pixel for a corresponding pixel,

Is the specific vector of the average moving-interest pixel,

Is the threshold signal difference,

Is the threshold distance,

Quot;

≪ / RTI >

Quot;

≪ / RTI >

Is the first vector component of the average vector V (X) calculated for X,

Is the second vector component of the mean vector V (X) calculated for X.)

Wherein the first distance is a value according to a norm of a difference between the first vector component of the average vector and the first vector component of the average moving pixel, Of the second vector component of the mean vector and a mean of a difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average motion- A second vector component of the second vector component of the average vector and a second vector component of the second vector component of the average vector, And is a value according to the norm of the difference between the two.

Here, if the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average motion-involved pixel is equal to or greater than a predetermined threshold value, Calculating the second distance by using a value obtained by multiplying the difference between the luminance signal of the second vector component of the average vector and the luminance signal among the second vector components of the average moving involved pixels by a predetermined weighting value .

Here, if the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average motion-involved pixel is equal to or greater than a predetermined threshold value, The difference between the texture channel signal of the second vector component of the average vector and the texture channel signal of the second vector component of the average motion-involved pixel is set to 0, and the second distance is calculated. can do.

Herein, the updating process is repeated until the absolute value of the difference between the average vectors before and after the updating becomes equal to or less than a predetermined value.

Here, the image dividing unit may divide the target image into partial images by integrating similar pixels having a predetermined difference between the average vectors and less than a predetermined reference, based on the average vector for each pixel of the target image, . ≪ / RTI >

According to another aspect of the present invention, there is provided an image segmentation method considering texture characteristics according to another aspect of the present invention. The image segmentation method includes receiving a target image to be segmented, A texture channel signal acquisition step of acquiring a texture channel signal representing a signal distribution characteristic of a regional image composed of pixels which are arranged in a predetermined direction; An average shifting process (Mean Shift Process) is performed for each pixel of the target image using the coordinate information of each pixel, the video signal information, and the obtained texture channel signal to obtain an average vector Process steps; And an image segmentation step of segmenting the target image into partial images using the average vector obtained in the average movement step.

Here, the acquiring of the texture channel signal may include acquiring a luminance image of the target image, setting an image block having a predetermined size for each pixel of the luminance image, and calculating a fractal dimension for the set image block And acquiring the texture channel signal using the calculated fractal dimension.

The average moving step may include obtaining a first vector component having a pixel coordinate value and a second vector component including a luminance signal, a color difference signal, and the texture channel signal, for each pixel of the target image, Selecting peripheral pixels within a threshold distance from each pixel for each pixel of the target image and having a threshold signal difference between the second vector component and each pixel, And calculates the average vector of each pixel using the average moving interest pixel.

Here, the average moving step may include using the first vector component and the second vector component of each pixel of the target image and the first vector component and the second vector component of the average motion-involved pixels for each pixel Wherein the average vector is calculated by calculating the average vector of each pixel, and the average vector is calculated through an iterative updating process, wherein the average vector is obtained by multiplying the first vector component, the luminance signal, And the second vector component including the color difference signal and the texture channel signal.

Calculating a first distance between the first vector component of the average vector and the first vector component of the average moving-participating pixel, wherein the first distance is calculated by multiplying the first component of the mean vector And a second vector component of the mean moving-participating pixel, wherein the second vector component of the average moving-involved pixel is a value according to a norm of a difference between the first vector of the average moving- Wherein the second distance is defined as a sum of a norm of a difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average moving- A norm of a difference between a chrominance signal of the second vector component and a chrominance signal of the second vector component of the average motion-involved pixel, and a norm of a difference between the texture channel signal and the texture channel signal of the second vector component of the average vector, Wherein the average vector is a value according to a norm of a difference between the texture channel signals of the second vector components of the gyro-motion-involved pixels, and updating the average vector using the first distance and the second distance .

In this case, if the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average motion-involved pixel is equal to or greater than a predetermined threshold value, The second distance is calculated using a value obtained by multiplying the difference between the luminance signal of the second vector component of the average vector and the latency signal of the second vector component of the average motion-involved pixel by a predetermined weighting value , And when the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average moving-participating pixel is equal to or greater than a predetermined threshold value, in calculating the second distance, The difference between the texture channel signal of the second vector component and the texture channel signal of the second vector component of the average motion-involved pixel is set to 0 It can be characterized by calculating the second distance.

According to the image dividing device and the method considering the reduction characteristics according to the present invention, it is possible to stably detect and distinguish the boundary between the flat area and the image of the texture, Is effectively divided into one area while suppressing the occurrence of over-segmentation.

1A and 1B are reference views showing an example of dividing an image using an image segmentation technique.
FIG. 2 is a block diagram of an image segmentation apparatus considering texture characteristics according to an embodiment of the present invention. Referring to FIG.
3 is a reference diagram for explaining the operation of the texture channel signal acquisition unit according to the present invention.
FIG. 4 is a reference diagram for explaining a method of varying the average vector calculation condition in the edge region in the average movement process unit according to the present invention.
FIGS. 5 and 6 are reference views for explaining an image segmentation result using an image segmentation device considering the texture characteristic according to the present invention, and solving the oversegmentation problem of the conventional image segmentation method.
7 is a flowchart of an image segmentation method considering texture characteristics according to another embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals are used to designate the same or similar components throughout the drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. In addition, the preferred embodiments of the present invention will be described below, but it is needless to say that the technical idea of the present invention is not limited thereto and can be variously modified by those skilled in the art.

An image segmentation technique that divides regions having similar properties in an image is a technique utilized in various technical fields such as object detection, object recognition, and partial image segmentation. Various techniques for dividing a partial image according to an object of an image have been proposed based on the edge. However, when the image is divided based on only the edge as described above, an over-segmentation problem occurs in which an area is excessively divided for an object having complicated texture properties. Therefore, an important point in such an image segmentation technique is to correctly discriminate regions having similar texture properties in the input image, and to identify regions.

FIG. 1A is a reference view showing an example in which an input image is divided using the image segmentation technique. Referring to FIG. 1A, it can be seen that FIG. 1A is divided according to objects included in an image. In general, image portions corresponding to a specific object have properties having the same or similar texture properties. Therefore, if the image is segmented according to the texture properties in the image, a meaningful partial image reflecting the position, structure and properties of the object can be obtained.

FIG. 1B shows that the texture of each object included in FIG. 1A has different properties. Referring to FIG. 1B, it can be seen that the sky portion T10 has a relatively smooth texture property, while the haystack portion T30 has a texture characteristic having a complicated pattern according to hay properties. The grass part (T20) is more complex than the sky part (T10), but has a texture property that is more flat than the hay dummy part (T30).

Therefore, there is a need to segment the image while considering the texture component in the image to be segmented.

In the present invention, a texture channel signal representing a complexity of a texture is calculated based on roughness of an image using a fractal image signal, and the image is divided on the basis of an average motion vector calculated using the texture channel signal. And the image boundary is stably detected and separated. In addition, a portion of the texture, which is determined by the same texture, is efficiently segmented into a single region while suppressing the occurrence of oversegmentation. Device and method therefor.

FIG. 2 is a block diagram of an image segmentation apparatus considering texture characteristics according to an embodiment of the present invention. Referring to FIG.

The image segmentation device considering the texture characteristic according to the present invention may include a texture channel signal acquisition unit 100, an average movement process unit 200, and an image segmentation unit 300.

The texture channel signal acquisition unit 100 receives a target image to be segmented and generates a texture channel signal representing a signal distribution characteristic of a local image including pixels positioned within a predetermined distance from the pixels of the target image, Signal.

The average movement process unit 200 performs a mean shift process for each pixel of the target image using the coordinate information of the pixels, the video signal information, and the obtained texture channel signal, And obtains a star average vector.

Here, the coordinate information means a coordinate where each pixel of the target image is located in the image. For example, when the upper left pixel of the target image is set as the origin, the coordinate information may be expressed by (x, y) which is a two-dimensional coordinate with respect to the origin. Here, the image signal information means luminance and chrominance signals in each pixel, and it is preferable to acquire luminance and chrominance signals by applying various color spaces as needed. Also, the video signal information may be a color signal such as R, G, or B, if necessary. That is, the video signal information is not limited to a specific color space and color conversion method, and can be acquired by applying various formats capable of expressing image signal components of each pixel.

Here, the mean shift process is a process of calculating and acquiring an average vector through an iterative update process. Here, the average vector is a vector calculated through a weighted average calculation of the specific vectors with respect to a specific vector set for each pixel of the object image. The operation of the average movement process unit 200 will be described in more detail below.

The image divider 300 divides the target image into partial images using the average vector obtained in the average movement processing unit 200. [

Herein, the image dividing apparatus considering the texture characteristic according to the present invention may further include a normalizing unit (not shown).

The normalization unit performs a normalization process of adjusting a dynamic range of the texture channel signal obtained from the texture channel signal acquisition unit 100 according to a dynamic range of a video signal of the target image, The average movement processing unit 200 may perform an average movement process using the texture channel signal normalized by the normalization unit. For example, if the dynamic range of the video signal of the target image is 0 to 255, the dynamic range of the texture channel signal can be adjusted accordingly.

Next, the operation of the texture channel signal acquisition unit 100 will be described in more detail.

The texture channel signal acquisition unit 100 calculates a variance of pixels located within a predetermined distance from each pixel of each pixel of the target image and sets the calculated variance to the texture channel signal of each pixel .

In addition, the texture channel signal acquisition unit 100 may set the texture channel signal to the fractal image obtained from the target image, as described below.

In this case, it is more preferable to set the texture channel signal to the fractal dimension image than to set the dispersion to the texture channel signal.

Preferably, the texture channel signal acquisition unit 100 acquires a fractal image by performing a local operation on each pixel of the target image, and sets the signal value of the fractal image as the texture channel signal value.

Here, the texture channel signal acquisition unit 100 acquires a luminance image of the target image, sets an image block having a predetermined size for each pixel of the luminance image, and performs a Fractal Dimension And the calculated fractal dimension may be used as the signal value of each pixel of the fractal image. Alternatively, the calculated fractal dimension may be normalized to a predetermined fixed number (for example, 0 to 255) to be a signal value for each pixel of the fractal image. Here, the fractal dimension can be calculated using a conventional DBC (Differential Box Counting) method as follows. Here, the fractal dimension image means an image in which the signal value of each pixel is the fractal dimension value. The texture channel signal acquisition unit 100 sets a square kernel for each pixel of the target image and calculates a fractal dimension of the partial image in the square kernel. Here, the fractal dimension is preferably a box counting dimension, which can be calculated using a conventional DBC (Differential Box Counting) method. Next, the fractal dimension calculated for each of the pixels may be a signal value of the fractal dimension image through a normalization process, which is mapped to a predetermined range (for example, 0 to 255).

The texture channel signal acquisition unit 100 sets an image block having a predetermined size around each pixel for each pixel of the target image, calculates the fractal dimension for the set image block, And the signal value of the fractal image is calculated.

3 is a reference diagram for explaining the operation of the texture channel signal acquisition unit 100 according to the present invention. 3 (a), the texture channel signal obtaining unit 100 sets image blocks having a predetermined size for each pixel and performs the above-described operation to obtain the fractal image as shown in FIG. 3 (b) .

Next, the operation of the average movement process unit 200 will be described in more detail.

As described above, the mean shift process is a process of calculating and acquiring an average vector through an iterative updating process. Here, the average vector is obtained by multiplying the specific vector set for each pixel of the target image, And is a vector calculated through a weighted average operation.

For this purpose, the average moving process may set a kernel function and calculate a weighted average operation by applying a weight calculated according to the kernel function to calculate the average vector.

That is, the average moving process can calculate the average vector by performing a weighted average operation as shown in Equation (1).

는 N(X)에 포함되는 화소들의 각 상기 특정 벡터이고, K는 상기 커널 함수이다.)(Where X (X) is the average vector, X is the specific vector in the pixel at which the average vector is calculated, and N (X) is the average value of the pixels located within a certain range centering on the pixel corresponding to X Each corresponding set of said specific vectors,

Is the specific vector of each of the pixels included in N (X), and K is the kernel function.

And, in the average moving process, it is possible to calculate the average vector by repeating the above procedure by setting the average vector calculated as Equation (1) to x.

Here, the average moving process unit 200 obtains, for each pixel of the target image, a first vector component having a pixel coordinate value and a second vector component consisting of a luminance signal, a color difference signal, and the texture channel signal .

와 같이 표현될 수 있다. 또한 상기 제2벡터 성분은 상기 대상 영상의 각 화소의 휘도 신호와 색차 신호와 상기 텍스쳐 채널 신호로 이루어지는 바,

와 같이 표현될 수 있다.(여기서 L은 휘도 신호, a, b는 색차 신호, T는 상기 텍스쳐 채널 신호를 지칭한다.)Wherein the first vector component is a coordinate value of each pixel of the target image,

Can be expressed as The second vector component includes a luminance signal, a color difference signal, and the texture channel signal of each pixel of the target image,

(Where L is a luminance signal, a and b are color difference signals, and T is the texture channel signal).

가 이하 상술할 바와 같이 상기 평균벡터를 산출하는 대상이 되는 특정 벡터가 된다. 이하에서 특정 벡터는 이와 같이 화소에서의 좌표 값과, 휘도 신호와 색차 신호와 상기 텍스쳐 채널 신호로 이루어지는 벡터를 지칭하기로 한다.Therefore, the average movement processing unit 200 calculates the first vector component represented by (x, y) and the second vector component represented by (L, a, b, T) for each pixel of the target image Can be obtained. The specific vector consisting of the first vector component and the second vector component thus obtained

Becomes a specific vector for which the average vector is to be calculated, as described below. Hereinafter, the specific vector will be referred to as a coordinate value in the pixel and a vector composed of the luminance signal, the color difference signal, and the texture channel signal.

Here, the average moving process unit 200 selects neighboring pixels within a threshold distance from each pixel for each pixel of the target image, the neighboring pixels having a difference between the second vector components and the threshold signal difference, And sets the average moving-involved pixel for each pixel, and calculates the average vector of each pixel using the average moving-involved pixel.

In other words, the average motion-involved pixel used in calculating the average vector is located within a threshold distance of a certain size from the pixel to be calculated as the average vector, and also, as described above (L, a, b, T ) Of the second vector component is within a threshold signal difference of a constant magnitude with respect to the second vector component.

The average motion-involved pixel may be selected as shown in Equation (2).

는 상기 평균이동 관여화소인지 여부의 판단 대상이 되는 화소의 특정 벡터이고,

는 상기 문턱신호차이이고,

는 상기 문턱거리이고,

는 상기 X의 상기 제1벡터 성분이고,

는 상기

의 상기 제1 벡터 성분이고,

는 상기 X의 상기 제2벡터 성분이고,

는 상기

의 상기 제1 벡터 성분이다.)(Where X is a specific vector in a target pixel from which the average vector is calculated and N (X) is a set of specific vectors of the average mobile-interest pixel for a pixel corresponding to the X,

Is a specific vector of a pixel to be judged as to whether or not it is the average motion-involved pixel,

Is the threshold signal difference,

Is the threshold distance,

Is the first vector component of X,

Quot;

≪ / RTI >

Is the second vector component of X,

Quot;

Of the first vector component.

Here, the average movement process unit 200 may calculate the first vector component and the second vector component of each pixel of the target image, and the first vector component and the second vector component of the average motion- , The average vector of each pixel is calculated.

In other words, the average movement processing unit 200 calculates the average vector of the target pixel by using the specific vector consisting of the first vector component and the second vector component of each pixel of the target image, It is preferable that the average vector of each pixel is calculated by using a specific vector composed of the second vector component.

Here, the average vector may be calculated using a specific vector of each pixel of the target image and a specific vector of the average motion-involved pixels for each pixel, as shown in Equation (3).

(Where the indices of the variables are the same as in Equation 2, V (X) is the average vector, and G is a kernel function for calculating the weight).

Here, the average movement processing unit 200 preferably calculates the average vector through an iterative updating process. That is, referring to Equation (3), the average vector V (X) calculated as Equation (3) is set to X, and the average vector calculation process is repeated as in Equation (3) The average vector can be finally calculated.

The initial updating of the average vector may be performed by setting the initial value of the average vector to a vector of the first vector component and the second vector component of each pixel, It is preferable to calculate the new average vector using the first vector component and the second vector component of the pixel and update the average vector.

In addition, the updating process after the first updating process updates the average vector by using the average vector updated in the previous updating process and the first vector component and the second vector component of the average moving involved pixel .

The initial update process may be performed as shown in Equation (4).

(Here, the display of the variables is the same as in Equation (3), and j is an index indicating the number of repetitions of the update process.)

Here, the average vector includes the first vector component having the pixel coordinate value, the luminance signal, the color difference signal, and the second vector component composed of the texture channel signal.

Here, the kernel function G for calculating a weight for calculating the average vector may be set as shown in Equation (5).

는 상기 X에 대하여 산출된 상기 평균벡터의 제1벡터 성분이고,

는 상기 X에 대하여 산출된 상기 평균벡터의 제2벡터 성분이다.)(Where the indication of the variables is the same as in Equation 4 above,

Is a first vector component of the mean vector calculated for X,

Is the second vector component of the mean vector calculated for X.)

Here, the average moving process unit 200 calculates a first distance between the first vector component of the average vector and the first vector component of the average moving-participating pixel, and calculates a second distance between the second vector component of the average vector Calculate a second distance between the second vector components of the average moving-interest pixels, and update the average vector using the first distance and the second distance.

Also, the average moving process unit 200 calculates the updated average vector using a value obtained by dividing the first distance by using the threshold distance and a value obtained by dividing the second distance by the threshold signal difference .

Here, the average movement processing unit 200 may calculate the average vector using Equation (4) and Equation (5), and the average vector calculation process may be expressed as Equation (6). Here, as described above, a vector including the first vector component and the second vector component in a pixel is referred to as a specific vector.

는 상기 평균이동 관여화소의 특정 벡터이고,

는 상기 문턱신호차이이고,

는 상기 문턱거리이고,

는 상기

의 상기 제1 벡터 성분이고,

는 상기

의 상기 제2 벡터 성분이고,

는 상기 X에 대하여 산출된 상기 평균벡터 V(X)의 제1벡터 성분이고,

는 상기 X에 대하여 산출된 상기 평균벡터 V(X)의 제2벡터 성분이다.)Wherein X is a specific vector in a target pixel from which the average vector is calculated, V (X) is the average vector, j is an index indicating the number of repetitions of the updating process, and N (X) A set of the specific vectors of the average motion-involved pixels for the pixels,

Is a specific vector of the average moving interest pixel,

Is the threshold signal difference,

Is the threshold distance,

Quot;

≪ / RTI >

Quot;

≪ / RTI >

Is the first vector component of the average vector V (X) calculated for X,

Is the second vector component of the mean vector V (X) calculated for X.)

Here, it is preferable that the updating process of the average moving process unit 200 is repeated until the absolute value of the difference between the average vectors before and after updating becomes less than a predetermined value.

Here, the first distance is preferably a value according to a norm of a difference between the first vector component of the average vector and the first vector component of the average motion-involved pixel, and can be calculated as follows: have.

는 상기 제1거리이고, 변수들의 표시는 상기 수학식 6과 동일하다.)(here

Is the first distance, and the indications of the variables are the same as in Equation (6).

And the second distance is calculated by multiplying the luminance signal among the second vector components of the average vector and the luminance signal among the second vector components of the average moving- And a second vector component of the second vector component of the average motion-involved pixel, a norm of a difference between a chrominance signal and a chrominance signal of the second vector component of the average motion- Is a value according to a norm of a difference between the texture channel signals of the second vector components. Here, the second distance may be calculated by Equation (8).

는 상기 제2거리이고, 변수들의 표시는 상기 수학식 6과 동일하고, L 은 상기 제2벡터 성분 중 휘도 신호, a와 b는 색차 신호, T는 상기 텍스쳐 채널 신호이다.)(here

Where a and b are chrominance signals, and T is the texture channel signal.) In the equation (6), L is a luminance signal among the second vector components, a and b are color difference signals, and T is the texture channel signal.

Here, the average moving process unit 200 may conditionally calculate the second distance using the luminance signal to improve the image segmentation performance in the edge region.

FIG. 4 is a reference diagram for explaining a method of varying the average vector calculation condition in the edge region in the average movement process unit according to the present invention. 4 (a) shows the target image with edges, and FIG. 4 (b) shows the texture channel signals obtained therefrom. Referring to FIG. 4, it can be seen that blurring appears in an edge portion of the texture channel signal. Such blurring at the edge portion deteriorates the image segmentation performance. Accordingly, as described below, the average movement process unit 200 changes the formula for calculating the second distance conditionally in the case of the edge portion of the target image, so that the image is robustly deformed according to the nature of the texture even at the edge portion .

For this, when the difference between the luminance signal among the second vector components of the average vector and the luminance signal among the second vector components of the average motion-involved pixels is equal to or greater than a predetermined threshold value, Calculating a second distance by using a value obtained by multiplying a luminance signal among the second vector components of the average vector and a norm of a difference between luminance signals in the second vector component of the average moving- , The second distance can be calculated.

If the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average motion-involved pixel is greater than a predetermined threshold value by the threshold signal difference, The second distance is calculated by using a value obtained by multiplying the difference between the luminance signal of the second vector component of the average vector and the luminance signal among the second vector components of the average motion involved pixel by a predetermined weight .

In this case, the average moving process unit 200 may calculate the second distance as shown in the following equation (9).

(Where T is the constant threshold, alpha is the constant weight, and the indications of the variables are the same as in Equation 6).

Here, the constant weight alpha may have a value of 2 to 4, and may be set to a different value if necessary.

Next, when the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average motion-involved pixel is equal to or greater than a predetermined threshold value, 2 distance, the difference between the texture channel signal of the second vector component of the average vector and the texture channel signal of the second vector component of the average motion-involved pixel is set to 0, and the second distance is calculated .

In this case, the average moving process unit 200 may calculate the second distance as shown in Equation (10).

(Where T is the constant threshold and the representation of the variables is the same as in Equation 6 above).

Here, the average movement processing unit 200 may calculate the second distance by simultaneously using the conditional second distance calculation of the two cases.

The image segmenting unit 300 integrates similar pixels having a predetermined difference between the average vectors and a reference pixel on a basis of the average vector for each pixel of the target image into one region, .

Here, the predetermined reference is a value that can be set according to the sensitivity or degree of the image segmentation unit 300 dividing the image.

Dividing the target image into partial images using the average vector may include merging pixels within a predetermined range of the difference between the average vectors on the basis of the average vector into one area, The partial image may be generated in such a manner that pixels exceeding a certain reference are classified into different regions. Here, in order to prevent a partial image from being too small, a minimum number of pixels per partial image is determined, and when the partial image is generated below the minimum number of pixels, it may be integrated with another partial image generated in the periphery.

Here, the image dividing unit 300 may use an existing method of dividing an image using an average vector.

FIGS. 5 and 6 are reference views for explaining an image segmentation result using an image segmentation device considering the texture characteristic according to the present invention, and solving the oversegmentation problem of the conventional image segmentation method. (B) is a result of image segmentation when the conventional image segmentation method is used, and (c) is a diagram illustrating an image segmentation result obtained for the segmentation target image according to the present invention. (D) is a result of dividing an image using an image dividing device considering a texture characteristic according to the present invention.

Referring to FIG. 5, in the case where a part having a complex texture is included in a divided object image as shown in FIG. 5 (a), the conventional image segmentation method includes an over- segmentation method in which an excessive image segmentation is performed in a complex part, A phenomenon occurs. However, the image segmentation device considering the texture characteristic according to the present invention is not limited to the above-mentioned case where the overhead segmentation phenomenon does not occur as in the result (d) of dividing the image through the average movement process using the texture channel signal as shown in (c) The results of the image segmentation are shown. This also applies to the case of FIG.

7 is a flowchart of an image segmentation method considering texture characteristics according to another embodiment of the present invention.

The image segmentation method considering the texture characteristics according to the present invention can operate in the same manner as the image segmentation apparatus considering the texture characteristics described in detail with reference to FIG. Therefore, overlapping portions will be briefly described below.

The image segmentation method considering the texture characteristic according to the present invention may include a texture channel signal acquisition step S100, an average movement step S200, and an image segmentation step S300.

The texture channel signal acquisition step (S100) receives a target image to be divided, and acquires a texture channel signal representing a signal distribution characteristic of a local image, which is composed of pixels positioned within a predetermined distance from the pixels, Signal.

In the average moving process step S200, a mean shift process is performed for each pixel of the target image using the coordinate information of the pixel, the video signal information, and the obtained texture channel signal, And obtains a star average vector.

In the image segmentation step S300, the target image is segmented into partial images using the average vector obtained in the average movement step S200.

The texture channel signal acquisition step S100 may include acquiring a luminance image of the target image, setting an image block having a predetermined size for each pixel of the luminance image, and performing a Fractal Dimension And obtain the texture channel signal using the calculated fractal dimension.

Here, in the average moving process step S200, for each pixel of the target image, a first vector component having a pixel coordinate value and a second vector component consisting of a luminance signal, a color difference signal, and the texture channel signal are obtained For each pixel of the target image, a neighboring pixel located within a threshold distance from each pixel and having a difference in the second vector component from the pixel is within a threshold signal difference, , And calculates the average vector of each pixel using the average moving-involved pixels.

Here, the average moving process step S200 may include the first vector component and the second vector component of each pixel of the target image and the first vector component and the second vector component of the average motion- It is preferable that the average vector of each pixel is calculated and the average vector is calculated through an iterative updating process.

Here, it is preferable that the average vector includes the first vector component having the pixel coordinate value, the luminance signal, the color difference signal, and the second vector component composed of the texture channel signal.

Here, the average moving process step S200 may include calculating a first distance between the first vector component of the average vector and the first vector component of the average moving-participating pixel, And a value according to a norm of a difference between the first vector and the first vector of the average motion-involved pixel.

The average moving process step S200 may further include calculating a second distance between the second vector component of the average vector and the second vector component of the average moving-participating pixel, 2 vector component and a luminance signal of the second vector component of the average moving-involved pixel, a color difference signal of the second vector component of the average vector, and a color difference signal of the average moving- A difference between the difference between the color channel signals in the second vector component and a difference between the texture channel signal and the texture channel signal in the second vector component of the average moving- And is a value according to a norm.

Here, the average moving process step (S200) preferably updates the average vector using the first distance and the second distance.

If the difference between the luminance signal among the second vector components of the average vector and the luminance signal among the second vector components of the average motion-involved pixels is equal to or greater than a predetermined threshold value, Calculating a second distance by using a value obtained by multiplying a difference between the luminance signal of the second vector component of the average vector and the latency signal of the second vector component of the average motion involved pixel by a predetermined weight, .

If the difference between the luminance signal among the second vector components of the average vector and the luminance signal among the second vector components of the average motion-involved pixels is equal to or greater than a predetermined threshold value, Calculating the second distance by setting the difference between the texture channel signal of the second vector component of the average vector and the texture channel signal of the second vector component of the average movement participating pixel to 0 in calculating the distance desirable. It is to be understood that the present invention is not limited to these embodiments, and all elements constituting the embodiment of the present invention described above are described as being combined or operated in one operation. That is, within the scope of the present invention, all of the components may be selectively coupled to one or more of them.

In addition, although all of the components may be implemented as one independent hardware, some or all of the components may be selectively combined to perform a part or all of the functions in one or a plurality of hardware. As shown in FIG. In addition, such a computer program may be stored in a computer readable medium such as a USB memory, a CD disk, a flash memory, etc., and read and executed by a computer to implement an embodiment of the present invention. As the recording medium of the computer program, a magnetic recording medium, an optical recording medium, a carrier wave medium, and the like can be included.

Furthermore, all terms including technical or scientific terms have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs, unless otherwise defined in the Detailed Description. Commonly used terms, such as predefined terms, should be interpreted to be consistent with the contextual meanings of the related art, and are not to be construed as ideal or overly formal, unless expressly defined to the contrary.

It will be apparent to those skilled in the art that various modifications, substitutions and substitutions are possible, without departing from the scope and spirit of the invention as disclosed in the accompanying claims. will be. Therefore, the embodiments disclosed in the present invention and the accompanying drawings are intended to illustrate and not to limit the technical spirit of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments and the accompanying drawings . The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents should be construed as falling within the scope of the present invention.

100: texture channel signal acquisition unit
200: average moving process part
300:
S100: Texture channel signal acquisition step
S200: Average moving process step
S300: Image segmentation step

Claims (24)

In the image dividing device,
A texture channel signal acquisition unit for acquiring a texture channel signal representing a signal distribution characteristic of a local image, the pixels being positioned within a predetermined distance from each pixel of each pixel of the target image, ;
An average shifting process (Mean Shift Process) is performed for each pixel of the target image using the coordinate information of each pixel, the video signal information, and the obtained texture channel signal to obtain an average vector Process part; And
And an image dividing unit for dividing the target image into partial images using the average vector obtained in the average moving process unit. The method according to claim 1,
Further comprising a normalization unit performing a normalization process of adjusting a dynamic range of the texture channel signal obtained by the texture channel signal acquisition unit according to a dynamic range of a video signal of the target image,
Wherein the average movement process unit performs an average movement process using the texture channel signal normalized by the normalization unit. The method according to claim 1,
Wherein the texture channel signal obtaining unit calculates a variance of signal values of pixels located within a predetermined distance from each pixel of each pixel of the target image and sets the calculated variance to the texture channel signal of each pixel Wherein the image processing unit is configured to process the texture data in accordance with the texture data. The method according to claim 1,
Wherein the texture channel signal obtaining unit obtains a fractal image by performing a local operation on each pixel of the target image and sets a signal value of the fractal image as the texture channel signal value. Considered image segmentation device. 5. The method of claim 4,
Wherein the texture channel signal obtaining unit obtains a luminance image of the target image, sets an image block of a predetermined size for each pixel of the luminance image, calculates a fractal size for the set image block, And calculating a signal value for each pixel of the fractal image using the calculated fractal dimension. 6. The method of claim 5,
Wherein the texture channel signal obtaining unit calculates a signal value for each pixel of the fractal image by normalizing the calculated fractal dimension to a predetermined number of predetermined ranges. 6. The method of claim 5,
And wherein the fractal dimension is a box counting dimension. The method according to claim 1,
Wherein the average moving process unit comprises:
Acquiring, for each pixel of the target image, a first vector component having a pixel coordinate value and a second vector component composed of a luminance signal, a color difference signal, and the texture channel signal,
Selecting peripheral pixels within a threshold distance from each pixel for each pixel of the target image and having a threshold signal difference between the second vector component and each pixel, and,
And calculates the average vector of each of the pixels using the average moving involved pixels. 9. The method of claim 8,
Wherein the average movement process unit uses the first vector component and the second vector component of each pixel of the target image and the first vector component and the second vector component of the average motion involved pixels for each pixel, Characterized in that the average vector of each pixel is calculated,
Wherein the average vector includes the first vector component having a pixel coordinate value, the luminance signal, the color difference signal, and the second vector component composed of the texture channel signal. 10. The method of claim 9,
Wherein the average moving process unit calculates the average vector through an iterative updating process. 11. The method of claim 10,
The initial updating process of the updating process sets an initial value of the average vector to a vector composed of the first vector component and the second vector component of each pixel, Calculating the new average vector using the first vector component and the second vector component, updating the average vector,
Updating the average vector using the first vector component and the second vector component of the average moving-participant pixel and updating the average vector in the previous updating process Characterized by a texture characteristic. 11. The apparatus according to claim 10,
Calculating a first distance between the first vector component of the average vector and the first vector component of the average mobile-
Calculating a second distance between the second vector component of the average vector and the second vector component of the average moving-participant pixel,
And the average vector is updated using the first distance and the second distance. delete 13. The method of claim 12,
Wherein the first distance is a value according to a norm of a difference between the first vector component of the average vector and the first vector component of the average moving-participant pixel,
Wherein the second distance is defined as a difference between a luminance signal of the second vector component of the average vector and a luminance signal of the second vector component of the average moving- A second vector component of the second vector component of the average motion-involved pixel, a norm of a difference between the color-difference signal and the second vector component of the average motion-involved pixel, 2 vector components are values according to a norm of the difference between the texture channel signals. 15. The method of claim 14,
The average moving process unit calculates the second distance when the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average moving-participating pixel is equal to or greater than a predetermined threshold value Wherein the second distance is calculated by using a value obtained by multiplying a difference between the luminance signal of the second vector component of the average vector and the luminance signal among the second vector components of the average motion involved pixel by a predetermined weight value. Which is based on texture characteristics. 15. The method of claim 14,
The average moving process unit calculates the second distance when the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average moving-participating pixel is equal to or greater than a predetermined threshold value Wherein the second distance is calculated by setting a difference between a texture channel signal of the second vector component of the average vector and a texture channel signal of the second vector component of the average motion- Image segmentation device considering texture characteristics. 11. The method of claim 10,
Wherein the update process is repeated until the absolute value of the difference between the average vectors before and after the update becomes equal to or less than a predetermined value. The method according to claim 1,
Wherein the image dividing unit divides the target image into partial images by integrating similar pixels having a difference between the average vectors that are equal to or less than a predetermined reference value into one area based on the average vector for each pixel of the target image, , And the texture characteristic is taken into account. In the image segmentation method,
A texture channel signal acquisition step of acquiring a texture channel signal representing a signal distribution characteristic of a local image, the input image being divided into pixels each of which is located within a predetermined distance from each pixel of each pixel of the object image; ;
An average shifting process (Mean Shift Process) is performed for each pixel of the target image using the coordinate information of each pixel, the video signal information, and the obtained texture channel signal to obtain an average vector Process steps; And
And an image segmentation step of segmenting the target image into partial images using the average vector obtained in the average moving process step. 20. The method of claim 19,
The texture channel signal acquisition step may include acquiring a luminance image of the target image, setting an image block having a predetermined size for each pixel of the luminance image, calculating a fractal size for the set image block, And acquiring the texture channel signal using the calculated fractal dimension. 20. The method of claim 19,
Wherein the average moving step comprises:
Acquiring, for each pixel of the target image, a first vector component having a pixel coordinate value and a second vector component composed of a luminance signal, a color difference signal, and the texture channel signal,
Selecting peripheral pixels within a threshold distance from each pixel for each pixel of the target image and having a threshold signal difference between the second vector component and each pixel, and,
And calculating the average vector of each of the pixels using the average moving-involved pixels. 22. The method of claim 21,
Wherein the average moving step comprises using the first vector component and the second vector component of each pixel of the target image and the first vector component and the second vector component of the average motion- The average vector of each pixel is calculated, and the average vector is calculated through an iterative updating process.
Wherein the average vector includes the first vector component composed of the pixel coordinate values, the luminance signal, the color difference signal, and the second vector component composed of the texture channel signal. 23. The method of claim 22,
Calculating a first distance between the first vector component of the average vector and the first vector component of the average moving-participant pixel, wherein the first distance is calculated by multiplying the first vector component of the mean vector and the And a value according to a norm of a difference between the first vector components,
Calculating a second distance between the second vector component of the average vector and the second vector component of the average moving-participant pixel, wherein the second distance is calculated by multiplying the luminance signal and the average movement A difference between the luminance signal difference of the second vector components of the associated pixel and the difference between the color difference signal of the second vector component of the average vector and the second vector component of the average moving- And a value according to a norm of a difference between the texture channel signal of the second vector component of the average vector and the texture channel signal of the second vector component of the average motion involved pixel, and,
Wherein the average vector is updated using the first distance and the second distance. 24. The method of claim 23,
Wherein when the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average motion-involved pixel is greater than or equal to a predetermined threshold value, Calculating the second distance by using a value obtained by multiplying the difference between the luminance signal of the second vector components and the luminance signal among the second vector components of the average moving-involved pixels by a predetermined weight,
Wherein when the difference between the luminance signal of the second vector component of the average vector and the luminance signal of the second vector component of the average motion-involved pixel is greater than or equal to a predetermined threshold value, Wherein the second distance is calculated by setting the difference between the texture channel signal of the second vector component and the texture channel signal of the second vector component of the average movement participating pixel to 0, Way.

Images