KR101296734B1 - Method for object segmentation by combination of gaussian mixture model and rgb clustering - Google Patents

Abstract

The present invention relates to a method for segmenting an object using a Gaussian mixture model and Algibi clustering. Specifically, when there is an object having an indefinite shape such as a flower in an image, a user automatically recognizes the object without a separate operation. And an object segmentation method using Gaussian mixture model and Algivy clustering that can be partitioned automatically.

Description

Method for object segmentation by combination of gaussian mixture model and rgb clustering}

The present invention relates to an object segmentation method using a Gaussian mixture model and Algibi clustering. Specifically, an object having an unspecified shape such as a flower in an image is automatically recognized and automatically recognized by the user without an additional manipulation. The present invention relates to a segmentable Gaussian mixture model and an object segmentation method using Algivy clustering.

In general, the image recognition field is a field for recognizing a specific object from an input image or an input image input from a photographing apparatus including a camera. Recently, a mobile terminal including a smart phone is commercially available and the camera is connected to the mobile terminal. As is basically provided, research and development for the user to recognize a particular object easily and conveniently using the mobile terminal is in progress.

In addition, in order to recognize the specific object, the object to be recognized in the input image or the input image should be divided. In this case, Gaussian mixture model (GMM), Markov random field (MRF), CV-Model, and K-average algorithm The object is partitioned using various algorithms such as k-means or importance map.

In addition, by using the mobile terminal equipped with the camera it is possible to recognize the object including the 'text', 'fingerprint' or 'poster' in the input image or the input image.

On the other hand, if the object to be recognized is made of an unspecified form such as a flower or a plant, and the background is complicated with various colors and shapes, the object to be recognized in the input image or the input image is accurately divided. There was a problem that could not be done.

In addition, since the division of the target object is not smooth and objects other than the target object are recognized as noise, there is a problem that the actual recognition is lowered and the accuracy is lowered.

In addition, in order to accurately divide an unspecified form of the target object, the algorithms require a process of drawing a contour of the target object directly by the user, which causes a problem of time consuming and cumbersome for the user. .

The present inventors have tried to make it possible to accurately recognize an object having an unspecified shape, such as a flower or a plant, and to automatically segment an object that is determined to be an object in an input image. The technical configuration of the object segmentation method used was developed to complete the present invention.

Accordingly, an object of the present invention is to provide a Gaussian mixture model and an object segmentation method using Algivy clustering that can automatically segment an unspecified object from an input image without a user's direct manipulation.

Another object of the present invention is to provide a Gaussian mixture model and an object segmentation method using Algivy clustering that can efficiently remove noise due to an unspecified object or background.

The objects of the present invention are not limited to the above-mentioned objects, and other objects not mentioned can be clearly understood by those skilled in the art from the following description.

In order to achieve the above object, the present invention provides an object segmentation method for dividing the object from the input image so that a computer apparatus can recognize an object in the input image, wherein the first sub-image generating means selects the computer apparatus from the input image. A first step of calculating normal distributions of each pixel to function to generate a first sub-image divided into a first object candidate region and a first background candidate region; The second sub-image generating means calculates the color average values of each candidate region and calculates the Euclidean distance between each pixel of the input image and each color average value, thereby generating the second object candidate region and the second background candidate. A second step of functioning to generate a second sub image divided into regions; And the object dividing means for generating the resultant image including the object dividing unit including the pixels located at coordinates overlapping each other of each pixel of the first object candidate region and each pixel of the second object candidate region. And a third step of functioning.

The present invention also provides a computer-readable storage medium storing an object partitioning program for performing the object partitioning method.

In a preferred embodiment, the noise removing means makes the computer apparatus binarize the resultant image to produce a binarized image in which the object segmentation area is divided into a plurality of binarization areas, the one having the largest area of each binarization area. And setting a binarization area as an object area and setting other areas as noise to function to remove the noise.

In a preferred embodiment, the object block generating means further comprises a fifth step of causing the computer device to generate, as an object block, pixels that match each pixel coordinate of the object region among the pixels of the input image. .

In a preferred embodiment, before the first step, a step A of causing the image format converting means to function the computer device to convert the format of the input image into an RGB color format.

In a preferred embodiment, the first step comprises: steps 1-1, wherein the first sub-image generating means functions to cause the computer apparatus to calculate the normal distributions using two Gaussian probability density functions; A first to second step of making a first sub image divided into divided regions having each normal distribution in the input image; And performing steps 1-3 to label a region having the lowest distribution density among the divided regions as a first object candidate region and to label another region as a first background candidate region.

In a preferred embodiment, the second step is the second sub-image generating means for causing the computer apparatus to calculate the respective RGB average values of the first object candidate region and the first background candidate region; step; A second step of calculating a RGB average value of the first object candidate region, an RGB average value of the first background candidate region, and an Euclidean distance value of each pixel of the input image; And labeling pixels having an Euclidean distance value close to the RGB average value of the first object candidate region among the pixels of the input image as the second object candidate region and other pixels as the second background candidate region. It includes; step 2-3 to function to generate a second sub-image.

In a preferred embodiment, the first sub-image generating means generates the first sub-image divided into the first object candidate region and the first background candidate region by using Equation 1 below. Function.

[Equation 1]

는 분포밀도이며, M은 가우시안 확률밀도 함수의 수이며,

는 평균벡터이며,

는 혼합 가중치(mixture weight)이며,

는 픽셀 x에 대한 확률밀도함수이다.here,

Is the distribution density, M is the number of Gaussian probability density functions,

Is the mean vector,

Is the mix weight,

Is the probability density function for pixel x.

In a preferred embodiment, the second sub image generating means functions to calculate the RGB average value of the first object candidate region by using Equation 2 below.

&Quot; (2) "

는 제 1오브젝트 후보영역의 RGB 평균값이며,

은 입력 이미지의 붉은(red) 색상값이고,

는 제 1오브젝트 후보영역의 붉은 색상 평균값이며,

는 입력 이미지의 녹색(green) 색상값이며,

는 제 1오브젝트 후보영역의 녹색 색상 평균값이며,

는 입력 이미지의 푸른(blue) 색상값이며,

는 제 1오브젝트 후보영역의 푸른 색상 평균값이다.here,

Is the RGB average value of the first object candidate region,

Is the red color value of the input image,

Is the average value of the red color of the first object candidate area,

Is the green color value of the input image,

Is the average value of the green color of the first object candidate area,

Is the blue color value of the input image,

Denotes a blue color average value of the first object candidate region.

In a preferred embodiment, the second sub image generating means functions to calculate the RGB average value of the first background candidate area by using Equation 3 below.

&Quot; (3) "

는 제 1배경 후보영역의 RGB 평균값이며,

은 입력 이미지의 붉은(red) 색상값이고,

는 제 1배경 후보영역의 붉은 색상 평균값이며,

는 입력 이미지의 녹색(green) 색상값이며,

는 제 1배경 후보영역의 녹색 색상 평균값이며,

는 입력 이미지의 푸른(blue) 색상값이며,

는 제 1배경 후보영역의 푸른 색상 평균값이다.here,

Is the RGB average value of the first background candidate area,

Is the red color value of the input image,

Is the average value of the red color of the first background candidate area,

Is the green color value of the input image,

Is the average value of the green color of the first background candidate area,

Is the blue color value of the input image,

Is the average value of the blue color of the first background candidate region.

In a preferred embodiment, the second sub image generating means functions to cause the computer apparatus to label each pixel of the input image as the second object candidate region or the second background candidate region using Equation 4 below. do.

&Quot; (4) "

는 제 1오브젝트 후보영역의 RGB 평균값이고

는 제 1배경 후보영역의 RGB 평균값이며,

는 제 2오브젝트 후보영역이며,

는 제 2배경 후보영역을 뜻한다.here,

Is the RGB average value of the first object candidate area

Is the RGB average value of the first background candidate area,

Is the second object candidate area,

Denotes a second background candidate region.

In a preferred embodiment, the object dividing means functions to cause the computer device to generate the object segment using an intersection algorithm.

In a preferred embodiment, the noise removing means functions to set the object area and the noise by the computer device using a connected component algorithm.

The present invention has the following excellent effects.

First, according to an object segmentation method according to an embodiment of the present invention, an intersection operation of each object candidate region of a first sub image segmented using a Gaussian probability density function and a second sub image segmented using RGB clustering is performed. By creating an object area, the object area of the target object to be recognized in the input image can be automatically divided without the user's direct manipulation, and the object area can be effectively divided even when the target object has an unspecified shape. You can get it.

In addition, according to the object segmentation method according to an embodiment of the present invention, since the noise removing means sets only a specific region as an object region and removes other regions, noise around the target object of an unspecified form and noise generated in the background are removed. The effect that can be removed efficiently can be obtained.

1 is a block diagram illustrating an object segmentation method according to an embodiment of the present invention.
2 is a view for explaining an object segmentation method according to an embodiment of the present invention.
3 to 4 is a view comparing the object segmentation method and the segmentation results of other algorithms according to an embodiment of the present invention.

Although the terms used in the present invention have been selected as general terms that are widely used at present, there are some terms selected arbitrarily by the applicant in a specific case. In this case, the meaning described or used in the detailed description part of the invention The meaning must be grasped.

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

However, the present invention is not limited to the embodiments described herein but may be embodied in other forms. Like reference numerals designate like elements throughout the specification.

The object dividing method according to an embodiment of the present invention is to divide the object from the input image and remove noise other than the object so that the computer device can recognize an object having an unspecified shape in the input image. Alternatively, the object may be divided so as to recognize an object having a specific shape such as flowers and plants without a specific shape such as a circle so as to be recognized in the input image.

In addition, the object segmentation method according to an embodiment of the present invention is substantially stored in the form of a program in a mobile terminal, PDA terminal or tablet PC equipped with a computer or a computing processor to perform a function.

In addition, the program is stored in a computer readable medium to perform the function by being read by a computer or mobile terminal, PDA terminal or tablet PC, the medium is those specially designed and configured for the present invention or computer software It may be known and available to those of ordinary skill in the art, for example, magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD, DVD, magnetic and optical recording A magnetic device, specially configured to store and execute the program instructions, alone or in combination, such as a magnetic-optical recording medium, ROM, RAM, flash memory, and the like.

In addition, the program is not only read by a computer, a mobile terminal, a PDA terminal or a tablet PC by the medium, but also stored in a server system capable of transmitting information through a communication network such as an intranet or the Internet, and then transmitted to a computer. Alternatively, the server system may provide a platform on which the computer may access the server system and execute the program on the server system without transmitting the program to the computer.

Hereinafter, an object division method according to an embodiment of the present invention will be described in detail.

1 is a block diagram illustrating an object segmentation method according to an exemplary embodiment of the present invention, and FIG. 2 is a diagram illustrating an object segmentation method according to an exemplary embodiment of the present invention.

1 to 2, in another object division method according to an embodiment of the present invention, first, when an input image photographed by a camera or a digital camera is input to a computer device, the image format converting means may be input by the computer device. By converting the format of the image, and also to generate the input image 110 converted to the RGB color format (S1100).

The computer device also includes a mobile terminal, PDA terminal or tablet PC equipped with a computer or computing processor.

The image format converting means may be a program or a means provided in the program functioning to allow the computer apparatus to convert the format of the input image.

Next, a first sub image is generated from the input image 110 converted to the RGB color format, wherein the first sub image is an image divided into a plurality of regions, and the first object candidate region 210 and the first background are generated. The candidate region 220 is divided (S1200).

In addition, the first sub image generating means may function to generate the first sub image by calculating the normal distributions of each pixel of the input image 110.

In addition, the first sub image generating means may be configured to divide the input image 110 in which the computer apparatus is converted into an RGB color format into the first object candidate region 210 and the first background candidate region 220. It may be a program or means provided in the program that functions to generate one sub-image.

In addition, the first sub-image generating means functions to calculate the normal distributions using a plurality of Gaussian probability density functions, and in one embodiment of the present invention, two normals using two Gaussian probability density functions. The distribution was calculated.

In addition, the first sub image generating means functions to generate the first sub image of the divided regions having the respective normal distributions, and the normal distribution of each of the divided regions has a different value. In addition, in an embodiment of the present invention, a first sub image divided into two divided regions is generated.

In addition, the first sub-image generating means functions to cause the computer apparatus to label the divided area into the first object candidate area 210 and the first background candidate area 220. If present, it is preferable to label the region having the lowest distribution density of each normal distribution as the first object candidate region 210 and to label another region as the first background candidate region 220.

In addition, in an embodiment of the present invention, a region having a low distribution density among two divided regions is labeled as the first object candidate region 210, and another region is labeled as the first background candidate region 220. It was.

In addition, the first sub-image generating means may be the first sub-image divided into the first object candidate region 210 and the first background candidate region 220 by using Equation 1 below. Function to generate

[Equation 1]

는 분포밀도이며, M은 가우시안 확률밀도 함수의 수이며,

는 평균벡터이며,

는 혼합 가중치(mixture weight)이며,

는 픽셀 x에 대한 확률밀도함수이다.here,

Is the distribution density, M is the number of Gaussian probability density functions,

Is the mean vector,

Is the mix weight,

Is the probability density function for pixel x.

That is, the number of divided regions divided in the input image 110 may be adjusted according to the number of Gaussian probability density functions.

Next, a second sub image is generated, and the second sub image generating means may be configured such that the computer apparatus determines each color average value of the first object candidate region 210 and the first background candidate region 220 and the input image ( The Euclidean distance between each pixel of 110 is calculated to generate the second sub image (S1300).

In addition, the second sub image is provided as an image divided into a second object candidate region 310 and a second background candidate region 320.

In addition, the second sub image generating means may be a program or a means included in the program functioning to cause the computer apparatus to generate the second sub image from the input image by using an RGB clustering algorithm.

In addition, the second sub image generating means functions to cause the computer apparatus to calculate respective RGB average values of the first object candidate region 210 and the first background candidate region 220.

In addition, the second sub image generating means may be configured such that the computer apparatus determines an RGB average value of the first object candidate region 210, an RGB average value of the first background candidate region 220, and each pixel of the input image 110. Functions to calculate Euclidean distance.

In addition, the second sub image generating means causes the computer apparatus to calculate an RGB average value of the first object candidate region 210 by using Equation 2 below, and the computer apparatus performs the following Equation 3 below. By using to calculate the RGB average value of the first background candidate region 220.

&Quot; (2) "

는 제 1오브젝트 후보영역의 RGB 평균값이며,

은 입력 이미지의 붉은(red) 색상값이고,

는 제 1오브젝트 후보영역의 붉은 색상 평균값이며,

는 입력 이미지의 녹색(green) 색상값이며,

는 제 1오브젝트 후보영역의 녹색 색상 평균값이며,

는 입력 이미지의 푸른(blue) 색상값이며,

는 제 1오브젝트 후보영역의 푸른 색상 평균값이다.here,

Is the RGB average value of the first object candidate region,

Is the red color value of the input image,

Is the average value of the red color of the first object candidate area,

Is the green color value of the input image,

Is the average value of the green color of the first object candidate area,

Is the blue color value of the input image,

Denotes a blue color average value of the first object candidate region.

&Quot; (3) "

는 제 1배경 후보영역의 RGB 평균값이며,

은 입력 이미지의 붉은(red) 색상값이고,

는 제 1배경 후보영역의 붉은 색상 평균값이며,

는 입력 이미지의 녹색(green) 색상값이며,

는 제 1배경 후보영역의 녹색 색상 평균값이며,

는 입력 이미지의 푸른(blue) 색상값이며,

는 제 1배경 후보영역의 푸른 색상 평균값이다.here,

Is the RGB average value of the first background candidate area,

Is the red color value of the input image,

Is the average value of the red color of the first background candidate area,

Is the green color value of the input image,

Is the average value of the green color of the first background candidate area,

Is the blue color value of the input image,

Is the average value of the blue color of the first background candidate region.

In addition, the second sub image generating means may be configured such that the computer apparatus includes pixels having the Euclidean distance value close to the RGB average value of the first object candidate region 210 among the pixels of the input image 110. The second object candidate region 310 is labeled and the other pixels are labeled as the second background candidate region 320.

In addition, the second sub image generating means may use the following Equation 4 to allow the computer apparatus to display each pixel of the input image 110 in the second object candidate region 310 or the second background candidate region 320. Label it).

&Quot; (4) "

는 제 1오브젝트 후보영역의 RGB 평균값이고

는 제 1배경 후보영역의 RGB 평균값이며,

는 제 2오브젝트 후보영역이며,

는 제 2배경 후보영역을 뜻한다.here,

Is the RGB average value of the first object candidate area

Is the RGB average value of the first background candidate area,

Is the second object candidate area,

Denotes a second background candidate region.

Next, the object dividing means causes the computer apparatus to generate a resultant image using the first object candidate region 210 and the second object candidate region 310, and the object dividing means causes the computer apparatus to generate the image. In operation S1400, an object partitioning region 410 including pixels located at coordinates overlapping each other among pixels of the first object candidate region 210 and each pixel of the second object candidate region 310 is performed.

Also. The result image includes the object segmentation area 410 and the background area 420, and the object division area 410 may be an area where a target object to be recognized in the input image is located and the background area 420 is located. May be an area other than the object partition area 410.

The object dividing means also functions to cause the computer device to generate the object dividing area 410 using an intersection algorithm. In addition, the intersection calculation algorithm may be an algorithm capable of generating an image by extracting pixels located at coordinates having the same characteristics among the pixels of the plurality of images and overlapping each other.

The second sub image generating means may generate the resultant image by performing an intersection operation on the first object candidate region 210 and the second object candidate region 310 using the intersection operation algorithm. It may be a functioning program or means provided in the program.

Next, the noise removing means functions to cause the computer device to remove noise included in the resultant image (S1500).

In addition, the noise removing means causes the computer apparatus to binarize the resultant image to generate a binarized image divided into a plurality of binarization regions, and the binarization region includes the object division region.

In addition, the noise removing means is configured to set one binarization area having the largest area among the binarization areas as the object area 510 and then set other areas 520 as noise to remove the noise area. . In this case, the binarization area set as the object area 510 has an area substantially overlapping or identical to the object partition area 410. The object may be accurately divided by removing the other area 520.

In addition, the noise removing means may be provided so that the computer device sets the object area 510 and the noise using a connected component algorithm.

The object block generating means then functions to cause the computer device to generate an object block in the input image, wherein the object block generating means causes the computer device to each pixel in the object area of each pixel of the input image. In operation S1600, the pixels matching the coordinates are generated as the object block.

In addition, the object block may consist of pixels positioned substantially on the outline of the object to be recognized, and may be utilized when the computer device recognizes the object.

3 to 4 are diagrams comparing object segmentation methods and segmentation results of other algorithms according to an exemplary embodiment of the present invention.

Referring to FIG. 3, an embodiment in which an object region and a background region are partitioned using an object segmentation method according to an embodiment of the present invention, and a first region in which an object region and a background region are partitioned using only a Gaussian mixture model (GMM) Comparative results were compared between the second comparative example of dividing the object region and the background region using only RGB clustering and the third comparative example of dividing the object region and the background region using a k-means algorithm.

In addition, the Examples and Comparative Examples used Intel (R) Core (TM) 2 Quad CPU Q9550 and MATLAB 7.7.0.171 (R2008b).

In addition, the input image used in the above examples and each comparative example was captured using the Internet or a mobile terminal, and a ground truth image having a flower and a background separated from each other was used for quantitative performance evaluation. In addition, the ground truth images were segmented by drawing flower areas directly in Adobe Photoshop CS4.

In addition, the division result of Comparative Example 1 shows that the object areas 10a and 10b do not correspond to the shape of the actual flower and appear indefinitely, and noises are generated in the background areas 11a and 11b.

In addition, the division result of Comparative Example 2 shows that object regions 20a and 20b including stems or leaves other than flowers are generated, and a plurality of noises are generated in the background regions 21a and 21b. .

In addition, it can be seen that the division result of Comparative Example 3 is also generated as the object areas 30a and 30b from the input image to areas other than the areas where flowers are located, and the background areas 31a and 31b also generate noise. .

On the other hand, the segmentation result generated by the object segmentation method according to an embodiment of the present invention, it can be seen that only the object region (100a, 100b) and the background region (110a, 110b) exist, no noise is generated, the object region It can be seen that (100a, 100b) coincides with the actual flower outline.

FIG. 4 shows the results obtained by comparing the above examples and respective comparative examples with Ground truth images and calculating the maximum, minimum, average, and standard deviations of similar degree using Equation 5 below.

&Quot; (5) "

Here, S is a similarity value, the closer to 1, the higher the similarity, trueforeground is the object region in the ground truth image, segmentedforeground is each object region in the above embodiment and each comparative example.

In addition, in the case of Comparative Example 1, it can be seen that the maximum value M2 of similarity is 0.9427 and the average value is relatively high as 0.83495, while the minimum value N2 is considerably low as 0.2951. That is, although the object region of Comparative Example 1 is substantially coincident with the Ground truth image, it can be seen that an almost inconsistent portion has also occurred.

In addition, in Comparative Example 2, the maximum value M3 of similarity was 0.9259 and the average value was 0.7813, which is relatively similar. However, the minimum value N3 of similarity is 0.3595, which is relatively low. That is, it can be seen that a part where the object area of Comparative Example 2 also does not substantially coincide with a certain part occurs.

In addition, in Comparative Example 3, it can be seen that the maximum value M4 of similarity is 0.901 and the average value is 0.7732, which is the lowest value of the comparative examples, and the minimum value N4 is 0.2009, which is the lowest value. That is, it can be seen that the object region of Comparative Example 3 is substantially the lowest in accuracy.

On the other hand, in the case of the embodiment, it can be seen that the maximum value M1 of the similarity has the highest value of 0.9539, the average value is 0.8815 the highest, and the minimum value N1 is 0.4568, which is higher than other comparative examples. That is, it can be seen that the object region of the embodiment is most similar to the ground truth image, and that the occurrence of mismatches is also least.

In addition, the standard deviation in the comparative examples and the embodiment is shown in the form of a box, which means that the shorter the vertical length and the smaller the size of the box, the smaller the standard deviation. In addition, comparing the standard deviations (S2, S3, S4) of the comparative example and the standard deviation (S1) of the embodiment it can be seen that the standard deviation (S1) of the embodiment is the smallest.

That is, the embodiment divided by the object segmentation method according to an embodiment of the present invention can be seen that the ground truth image is the most similar, and reflects the actual flower area.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is clearly understood that the same is by way of illustration and example only and is not to be taken by way of limitation in the present invention. Various changes and modifications will be possible.

Claims (13)

An object dividing method for dividing the object in the input image so that a computer device can recognize an object in the input image.
A first step of causing the computer apparatus to generate a first sub-image divided into a first object candidate region and a first background candidate region by calculating a normal distribution of each pixel of the input image by the first sub image generating means; ;
The second sub-image generating means calculates the color average values of each candidate region and calculates the Euclidean distance between each pixel of the input image and each color average value, thereby generating the second object candidate region and the second background candidate. A second step of functioning to generate a second sub image divided into regions; And
And the object dividing means causes the computer apparatus to generate a resultant image including an object segmentation region comprising pixels located at coordinates overlapping each other of each pixel of the first object candidate region and each pixel of the second object candidate region. And a third step of causing the object to be divided.
The method of claim 1,
The noise removing means binarizes the resultant image to generate a binarized image in which the object partition region is divided into a plurality of binarization regions, and converts one binarization region having the largest region among the binarization regions into an object region. And setting the other areas as noise and functioning to remove them.
The method of claim 2,
And a fifth step of generating, by the object block generating means, the computer device to generate, as an object block, pixels that match each pixel coordinate of the object area among the pixels of the input image. Way.
The method of claim 1,
Before the first step,
And an image format converting means for causing the computer device to function to convert the format of the input image into an RGB color format.
The method of claim 1,
The first step is
The first sub image generating means is the computer device,
1-1 to function to calculate the normal distributions using two Gaussian probability density functions;
A first to second step of making a first sub image divided into divided regions having each normal distribution in the input image; And
And labeling a region having the lowest distribution density among the divided regions as a first object candidate region, and performing another function of labeling another region as a first background candidate region. Division method.
The method of claim 1,
The second step is
The second sub image generating means is configured to
A step 2-1 of calculating a mean value of RGB values of the first object candidate region and the first background candidate region;
A second step of calculating a RGB average value of the first object candidate region, an RGB average value of the first background candidate region, and an Euclidean distance value of each pixel of the input image; And
Among the pixels of the input image, pixels having an Euclidean distance value close to an RGB average value of the first object candidate region are labeled as the second object candidate region, and other pixels are labeled as a second background candidate region, thereby forming the first object candidate region. And a second step of allowing the second sub image to be generated as the second sub image.
6. The method of claim 5,
The first sub image generating means functions to generate the first sub image divided into the first object candidate region and the first background candidate region by using Equation 1 below. How to split an object.
[Equation 1]

here,

Is the distribution density, M is the number of Gaussian probability density functions,

Is the mean vector,

Is the mix weight,

Is the probability density function for pixel x.
The method according to claim 6,
And the second sub image generating means functions to calculate the RGB average value of the first object candidate region by using Equation 2 below.
&Quot; (2) "

here,

Is the RGB average value of the first object candidate region,

Is the red color value of the input image,

Is the average value of the red color of the first object candidate area,

Is the green color value of the input image,

Is the average value of the green color of the first object candidate area,

Is the blue color value of the input image,

Denotes a blue color average value of the first object candidate region.
The method according to claim 6,
And the second sub image generating means functions to calculate the RGB average value of the first background candidate region by using Equation 3 below.
&Quot; (3) "

here,

Is the RGB average value of the first background candidate area,

Is the red color value of the input image,

Is the average value of the red color of the first background candidate area,

Is the green color value of the input image,

Is the average value of the green color of the first background candidate area,

Is the blue color value of the input image,

Is the average value of the blue color of the first background candidate region.
The method according to claim 6,
And the second sub image generating means functions to label each pixel of the input image as the second object candidate region or the second background candidate region by using Equation 4 below. Division method.
&Quot; (4) "

here,

Is the RGB average value of the first object candidate area

Is the RGB average value of the first background candidate area,

Is the second object candidate area,

Denotes a second background candidate region.
The method of claim 1,
And the object dividing means functions to cause the computer device to generate the object segmentation area using an intersection algorithm.
The method of claim 2,
And said noise removing means functions to set said object area and said noise by said computer device using a connected component algorithm.
A computer-readable storage medium storing an object partitioning program for performing the object partitioning method by functioning as a means of any one of claims 1 to 12.

Images