Image segmentation method, device and apparatus

Technical Field

The invention relates to a technical field of Image processing, in particular an Image segmentation method, device and apparatus.

Background Art

With the rapid growth of network multimedia information, the head shoulder and the segmentation technique, as a kind of special Image segmentation technique of a catalytic development. It has now been widely applied to such as video Conference background replacement, pats the human front camera on a mobile device, such as the rear camera [...] in the scene.

In the related art, in the Image while cutting apart the head and shoulder, first indicate to the user selecting the head shoulder and the prospect of the sample picture element spot and background of the 1st 2nd sample pixel point; after, respectively calculating the sample pixel point 1st and 2nd sample pixel point of the color feature vector, to obtain the prospects of shoulder 1st color feature vector and the background of the 2nd color feature vector; according to the 1st and 2nd color feature vector color feature vector, respectively correct shoulder prospect and the background color modeling, 1st and 2nd obtained color model color model; finally, the 1st and 2nd color model color model to the picture the head shoulder and the segmentation, the head shoulder and the segmentation result obtained.

In the realization of the process of the present disclosure, the inventor to find the relevant technical at least the following problems:

In the Image segmentation process need user participation in the selection of the sample pixel point, so the user experience is worse; in addition, since only the color of the sample pixel point based on the feature vector carrying out Image segmentation, so segmentation accuracy is poor.

Content of the invention

In order to overcome the problems existing in the related technology, the present invention provides an Image segmentation method, device and apparatus.

According to the disclosed embodiment of the 1st aspect, provides an Image segmentation method, said method comprising:

According to face the external outline of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability;

According to the a priori probability, the pre-set shoulder prospect probability threshold and background probability threshold, to be in split Image selecting the head shoulder and the sample picture element spot and the prospects for the background sample pixel point;

According to the prospects [...] sample pixel point of the color feature vector and the background sample pixel point of the color feature vector, calculating the 1st [...] prospect of color likelihood probability and the background of the 2nd color likelihood probability;

According to the a priori probability, the 1st color likelihood probability and the 2nd color likelihood probability, calculated by the 1st [...] prospect of the posteriori probability and the background of the 2nd a posteriori probability;

According to the 1st a posteriori probability and the 2nd a posteriori probability, the to-be-divided Image divided head and shoulder.

Optionally, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, including:

Selecting a preset number of female human face picture of front site;

According to the user for each site of female human face picture of front shoulder prospect calibration result, generates a piece of female human face picture of front shoulder of the calibration Image;

For each piece of female front human face Image face external contour characteristic point positioning, the positioning result obtained;

According to face the exterior outline of the positioning result of feature points, each of the shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image;

For multiple preset size in the Image of the same position pixel point, according to the prospects [...] calibration result, calculating the pixel point position where the head shoulder and the prospect of a priori probability.

Optionally, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, including:

The site selected preset number of male front human face Image;

According to the user for each site of the male human face picture of front shoulder prospect calibration result, generates a piece of male human face picture of front shoulder of the calibration Image;

For each piece of male front human face Image face external contour characteristic point positioning, the positioning result obtained;

According to the external outline of the face of the positioning result, the shoulder of each site to a calibration Image alignment and size normalization processing, get the multiple preset size Image;

For multiple preset size in the Image of the same position pixel point, according to the prospects [...] calibration result, calculating the pixel point position where the head shoulder and the prospect of a priori probability.

Optionally, the according to the prospects [...] calibration result, use the following formula, to calculate the pixel point position where the head shoulder and the prospect of a priori probability:

Wherein Refer to pixel point (x_i , Y_o ) Present at the head shoulder and the prospect of a priori probability, Refer to section j of images pixel point (x_i , Y_o ) At the calibration result, J site of said pixel in the Image point (x_i , Y_o ) Calibrated at [...] prospect, J site of said pixel in the Image point (x_i , Y_o ) At the calibration into background, N the front side of the indicating the number of the face Image.

Optionally, the according to the a priori probability, the pre-set shoulder prospect probability threshold and background probability threshold, to be in split Image selecting the head shoulder and the sample picture element spot and the prospects for the background sample pixel point, including:

The to-be-split Image of the human face in the gender identification, gender identification result obtained;

According to the results of the gender identification, determined to be divided Image corresponding to the priori probability;

For the to-be-in split Image of each pixel point, judging whether the pixel point is greater than the a priori probability [...] prospect probability threshold; if the pixel point of the a priori probability is greater than the probability threshold [...] prospects, will be the pixel point as foreground sample pixel point;

For the to-be-in split Image of each pixel point, judging whether the pixel point of the a priori probability is less than the background probability threshold; if the pixel point of the a priori probability is less than the background probability threshold, then the pixel point as background sample pixel point.

Optionally, the [...] prospects according to the color of the sample pixel point feature vector and the background sample pixel point of the color feature vector, use the following formula, calculating the 1st [...] prospect of color likelihood probability and the background of the 2nd color likelihood probability, including:

Wherein Refer to section j clusters of central color feature vector, Refer to 1st color likelihood probability, 2nd indicating color likelihood probability, N_F To refer to the head shoulder and the prospect sample pixel point of the number of the cluster center, N_B To refer to the background sample pixel point of the number of the cluster center; Refer to color feature vector To the cluster center European distance, w_j Refer to section j clusters in the center of the sample pixel point of the total sample pixel point ratio, β refer to different cluster of central color feature vector between the average European distance value, it is defined as:

Optionally, the according to the a priori probability, the 1st color likelihood probability and the 2nd color likelihood probability, the application of the following formula, calculate the posterior probability of the [...] prospects, the background of the posterior probability, including:

Wherein To refer to the posterior probability of the [...] prospects;

To refer to the background of the posterior probability.

According to the invention of the embodiment of in the 2nd, provides an Image dividing device, the device including:

The a priori probability calculation module, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability;

Sample pixel point selecting module, for according to the a priori probability, the pre-set shoulder prospect probability threshold and background probability threshold, to be in split Image selecting the head shoulder and the sample picture element spot and the prospects for the background sample pixel point;

Color likelihood probability calculating module, for according to the prospects [...] sample pixel point of the color feature vector and the background sample pixel point of the color feature vector, calculating the 1st [...] prospect of color likelihood probability and the background of the 2nd color likelihood probability;

A posteriori probability calculation module, for according to the a priori probability, the 1st color likelihood probability and the 2nd color likelihood probability, calculated by the 1st [...] prospect of the posteriori probability and the background of the 2nd a posteriori probability;

Image segmentation module, according to the 1st a posteriori probability and the 2nd a posteriori probability, the to-be-divided Image divided head and shoulder.

Optionally, the a priori probability calculating module, and is used for selecting a preset number of female human face picture of front site; according to the user for each site of female human face picture of front shoulder prospect calibration result, generates a site of female human face picture of front shoulder calibration Image; a site to each female front human face Image face external contour characteristic point positioning, the positioning result obtained; according to face the exterior outline of the positioning result of feature points, each of the shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image; for one of a plurality of preset size in the Image of the same position pixel point, according to the prospects [...] calibration result, calculating the pixel point position where the head shoulder and the prospect of a priori probability.

Optionally, the a priori probability calculating module, and is used for selecting a preset number of male human face picture of front site; according to the user for each site of the male human face picture of front shoulder prospect calibration result, generates a site of male human face picture of front shoulder calibration Image; a site to each male front human face Image face external contour characteristic point positioning, the positioning result obtained; according to face the exterior outline of the positioning result of feature points, the shoulder of each site to a calibration Image alignment and size normalization processing, get the multiple preset size Image; for one of a plurality of preset size in the Image of the same position pixel point, according to the prospects [...] calibration result, calculating the pixel point position where the head shoulder and the prospect of a priori probability.

Optionally, the a priori probability calculating module, the application of the following formula, to calculate the pixel point position where the head shoulder and the prospect of a priori probability:

Wherein Refer to pixel point (x_i , Y_o ) Present at the head shoulder and the prospect of a priori probability, Refer to section j of images pixel point (x_i , Y_o ) At the calibration result, J site of said pixel in the Image point (x_i , Y_o ) Calibrated at [...] prospect, J site of said pixel in the Image point (x_i , Y_o ) At the calibration into background, N the front side of the indicating the number of the face Image.

Optionally, said sample pixel point selecting module, used for the to-be-split Image of the human face in the gender identification, gender identification result obtained; according to the results of the gender identification, determined to be divided Image corresponding to the priori probability; for the to-be-in split Image of each pixel point, judging whether the pixel point is greater than the a priori probability [...] prospect probability threshold; if the pixel point of the a priori probability is greater than the probability threshold [...] prospects, will be the pixel point as foreground sample pixel point; for the to-be-in split Image of each pixel point, judging whether the pixel point of the a priori probability is less than the background probability threshold; if the pixel point of the a priori probability is less than the background probability threshold, then the pixel point as background sample pixel point.

Optionally, the color likelihood probability calculating module, the application of the following formula, calculating the 1st [...] prospect of color likelihood probability and the background of the 2nd color likelihood probability:

Wherein Refer to section j clusters of central color feature vector, Refer to 1st color likelihood probability, 2nd indicating color likelihood probability, N_F To refer to the head shoulder and the prospect sample pixel point of the number of the cluster center, N_B To refer to the background sample pixel point of the number of the cluster center; Refer to color feature vector To the cluster center European distance, w_j Refer to section j clusters in the center of the sample pixel point of the total sample pixel point ratio, β refer to different cluster of central color feature vector between the average European distance value, it is defined as:

Alternatively, the posterior probability calculating module, the application of the following formula, calculate the posterior probability of the [...] prospects, the background of the posterior probability, including:

Wherein To refer to the posterior probability of the [...] prospects;

To refer to the background of the posterior probability.

According to the invention of the embodiment of in the 3rd, provides an Image segmentation apparatus, the apparatus including:

A processor;

For storing processor executable instruction memory;

Wherein the processor is configured to: according to face the external outline of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability; according to the a priori probability, the pre-set shoulder prospect probability threshold and background probability threshold, to be in split Image selecting the head shoulder and the sample picture element spot and the prospects for the background sample pixel point; according to the prospects [...] sample pixel point of the color feature vector and the background sample pixel point of the color feature vector, calculates the 1st [...] prospect of color likelihood probability and the background of the 2nd color likelihood probability; according to the a priori probability, the 1st color likelihood probability and the 2nd color likelihood probability, calculated by the 1st [...] prospect of the posteriori probability and the background of the posteriori probability 2nd; according to said 1st a posteriori probability and the 2nd a posteriori probability, the to-be-divided Image divided head and shoulder.

The disclosed embodiment provides a technical scheme can include the following beneficial effects:

According to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, based on the a priori probability of the prospects for the shoulder of the head shoulder and the prospects for automatically selecting the sample picture element spot and background sample pixel point, to reach in the sample pixel point selection process is not needed in the involvement of the user, the user experience is good; in addition, in the Image segmentation based on the a priori probability of the color and the likelihood probability divided, segmentation criteria are more detailed, segmentation accuracy is high.

It should be understood that, the above general description of the details of the latter article and described is only illustrative and explanatory, does not limit the present invention.

Description of drawings

Here in the specification of the Figure are incorporated into and form a part of this specification, are shown in accordance with the disclosed embodiments, and together with the description used for explaining the principle of the disclosure.

Figure 1 is the flow chart of an exemplary implementation instantiates according to a kind of Image segmentation method.

Figure 2 is the flow chart of an exemplary implementation instantiates according to a kind of Image segmentation method.

Figure 3 is the block diagram of an exemplary implementation instantiates according to a kind of Image segmentation device.

Figure 4 is the block diagram of an exemplary implementation instantiates according to a kind of Image segmentation apparatus.

Figure 5 is the block diagram of an exemplary implementation instantiates according to a kind of Image segmentation apparatus.

Mode of execution

Here will be in detail to the description of the illustrative embodiment, the example that in the attached drawing. The following description relates to Figure when, unless otherwise expressed, different with photos in the same digital to represent the same or similar elements. The following exemplary embodiment described in the embodiment do not represent all of the with the present invention consistent with embodiments. On the contrary, they are only with such as the attached claims set forth in detail, of the present disclosure consistent with some aspects of the apparatus and method of the example.

Figure 1 is the flow chart, as shown in Figure 1, the Image segmentation method for graph of according to one exemplary implementation instantiates a Image segmentation method in Image segmentation apparatus, comprises the following steps.

In step 101 in, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability.

In step 102 in, according to the a priori probability, the pre-set shoulder prospect probability threshold and background probability threshold, to be in split Image selecting the head shoulder and the sample picture element spot and the prospects for the background sample pixel point.

In step 103 in, according to the prospects for shoulder sample pixel point feature vector and the color of the background sample pixel point of the color feature vector, calculating the head shoulder and the prospect of color likelihood probability and 1st 2nd background of color likelihood probability.

In step 104 in, according to the a priori probability, 1st and 2nd color color likelihood probability likelihood probability, calculating the head shoulder and the prospect of posterior probability and the 1st 2nd background of the posteriori probability.

In step 105 in, according to the 1st and 2nd of the posterior probability of the posteriori probability, treatment of dividing Image divided head and shoulder.

The disclosed embodiment of the method, according to the face due to the external outline of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, based on the a priori probability of the prospects for the shoulder of the head shoulder and the prospects for automatically selecting the sample picture element spot and background sample pixel point, to reach in the sample pixel point selection process is not needed in the involvement of the user, the user experience is good; in addition, in the Image segmentation based on the a priori probability of the color and the likelihood probability divided, segmentation criteria are more detailed, segmentation accuracy is high.

Optionally, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, including:

Selecting a preset number of female human face picture of front site;

According to the user for each site of female human face picture of front shoulder prospect calibration result, generates a piece of female human face picture of front shoulder of the calibration Image;

For each piece of female front human face Image face external contour characteristic point positioning, the positioning result obtained;

According to face the exterior outline of the positioning result of feature points, each of the shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image;

For multiple preset size in the Image of the same position pixel point, according to the prospects for shoulder calibration result, calculating the point position where the head shoulder and the prospect of a priori probability.

Optionally, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, including:

The site selected preset number of male front human face Image;

According to the user for each site of the male human face picture of front shoulder prospect calibration result, generates a piece of male human face picture of front shoulder of the calibration Image;

For each piece of male front human face Image face external contour characteristic point positioning, the positioning result obtained;

According to face the exterior outline of the positioning result of feature points, each of the shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image;

For multiple preset size in the Image of the same position pixel point, according to the prospects for shoulder calibration result, calculating the point position where the head shoulder and the prospect of a priori probability.

Optionally, the prospects for the shoulder according to the calibration results, use the following formula, calculating the point position where the head shoulder and the prospect of a priori probability:

Wherein Refer to pixel point (x_i , Y_o ) Present at the head shoulder and the prospect of a priori probability, Refer to section j of images pixel point (x_i , Y_o ) At the calibration result, J site of said pixel in the Image point (x_i , Y_o ) Calibrated at [...] prospect, J site of said pixel in the Image point (x_i , Y_o ) At the calibration into background, N the front side of the indicating the number of the face Image.

Optionally, according to the a priori probability, the pre-set shoulder prospect probability threshold and background probability threshold, to be in split Image selecting the head shoulder and the sample picture element spot and the prospects for the background sample pixel point, including:

Treatment of the human face in split Image gender identification, gender identification result obtained;

According to the results of the gender identification, determining corresponding to the a priori probability of the segmented Image;

For each of the to be in split Image a pixel point, judging the pixel point is greater than the a priori probability of the shoulder prospect probability threshold; if the pixel point of the a priori probability is greater than shoulder prospects probability threshold, then the pixel point as prospect sample pixel point;

For each of the to be in split Image a pixel point, judging the pixel point of the a priori probability is less than the background probability threshold; if the pixel point of the a priori probability is less than the background probability threshold, then the pixel point as background sample pixel point.

Optionally, according to the prospects for shoulder sample pixel point feature vector and the color of the background sample pixel point of the color feature vector, use the following formula, calculating the head shoulder and the prospect of color likelihood probability and 1st 2nd background of color likelihood probability, including:

Wherein Refer to section j clusters of central color feature vector, Refer to 1st color likelihood probability, 2nd indicating color likelihood probability, N_F To refer to the head shoulder and the prospect sample pixel point of the number of the cluster center, N_B To refer to the background sample pixel point of the number of the cluster center; Refer to color feature vector To the cluster center European distance, w_j Refer to section j clusters in the center of the sample pixel point of the total sample pixel point ratio, β refer to different cluster of central color feature vector between the average European distance value, it is defined as:

Optionally, according to the a priori probability, 1st and 2nd color color likelihood probability likelihood probability, the application of the following formula, calculate the posterior probability of the head shoulder and the prospects, the posterior probability of the background, including:

Wherein To refer to the head shoulder and the prospect of posterior probability;

Background indicating the posterior probability.

All of the above optional technical scheme, may employ any combined to form an alternative embodiment of the present disclosure, will not re-enumerate.

Figure 2 is the flow chart, as shown in Figure 2, the Image segmentation method for graph of according to one exemplary implementation instantiates a Image segmentation method in Image segmentation apparatus, comprises the following steps.

In step 201 in, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability.

In the disclosed embodiment, the prospects for the shoulder of the computation process of the a priori probability, also the establishment of the a priori model of the head shoulder and the position of the process. Because the male and female face of the characteristics of difference, so in the establishment of the a priori model when the position of the head and shoulder, also are required to establish the position of the female head and shoulder a priori model and the position of the male head and shoulder a priori model. Wherein the position of the head and shoulder a priori model of the establishing process are as follows:

201 A, the position of the female head and shoulder a priori model of the establishing process are divided into the following five steps.

1st step, selecting a preset number of female front human face Image site.

Wherein the size of the preset number is 500 or 1000 and so on, the disclosed embodiment is not limited to the specific. The preset number of female front human face Image as the site to obtain the position of the follow-up of the female head and shoulder a priori model training data stored in the Image training in a library.

2nd step, according to the user for each site of female human face picture of front shoulder prospect calibration result, generates a site of female human face picture of front shoulder calibration Image.

In the process in the off-line training, the preset number of female front human face Image of the site shoulder prospect calibration, the need for manpower to finish. Namely, requiring the user to manually calibration every single piece of the female front human face Image in the region of the head and shoulder. In the get every piece of female front human face Image calibration results, generating a binary images, and the two-value Image as the head shoulder and the calibration Image. Wherein in the binary pixel gray level 255 of the prospects for the head and shoulder region, that is the two-value Image in the white area of the for the head and shoulder region; in the binary gray pixel point 0 of the area as the background, that is the two-value Image in the black region as the background area.

In addition, a female front human face Image calibration when the head shoulder and the prospect, for a pixel point, if it is the prospect of shoulder, can their identification garneting 1, if it is a background, can their identification garneting 0. In order to pixel point (x_i , Y_i ) As an example, is To represent a pixel point (x_i , Y_i ) Calibration [...] prospect, To represent a pixel point (x_i , Y_i ) Calibrated into the background.

3rd step, for each piece of female front human face Image face external contour characteristic point positioning, the positioning result obtained.

In the disclosed embodiment, each site in the female front human face images when the face external contour characteristic localization, can be taken of the feature points of the face the exterior profile of the existing positioning algorithm to realize, in this does not then repeat. A female front human face Image of the human face the exterior profile of the feature points after the positioning, can be each a human face the exterior profile of the position coordinate data of the characteristic points.

4th step, according to the external outline of the face of the positioning result, each shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image.

Wherein the pre-set can have a 400 * 400,200 * 200 and the like, the disclosed embodiment of this are not specifically limited. The present invention the execution to the preset size is 400 * 400 is an example for illustration.

In the disclosed embodiment, for each shoulder calibration Image for a site, the obtained external outline of feature points of the face after the positioning result, according to the first face of the external outline of the head and shoulder the positioning result calibration Image to adjust to a uniform size; after, according to the feature points of the face the exterior profile of the position coordinate data of the calibration Image with the head and shoulder face the external outline of the aligning, and the head and shoulders calibration Image to normalize to a 400 * 400 in the template. Since each of the head and shoulder calibration images have corresponding to a 400 * 400 of the template, so can get the multiple preset size Image.

5th step, for multiple preset size in the Image of the same position pixel point, according to the prospects for shoulder calibration result, calculating the point position where the head shoulder and the prospect of a priori probability.

In the disclosed embodiment, the plurality of pieces of preset size in the Image of the same position pixel point, the pixel position where the head shoulder shoulder prospect of the a priori probability, i.e. each Image training library in a piece of female front human face Image in the pixel point is calibration [...] prospect of frequency. For example, the assumption that 1000 pieces of female front human face Image, a pixel point in 500 pieces of female front human face in the Image calibration [...] prospect, the pixel point is the position of the head and shoulder of the emergence of the prospects for the priori probability is 0.5. Therefore, the statistics of each pixel point in the training data in the frequency of the calibration [...] prospects, can then be obtained for each pixel point position where the probability of the head shoulder and the prospects.

201 B, the position of the male head and shoulder a priori model of the establishing process are divided into the following five steps.

1st step, selecting a preset number of male site front human face Image.

Wherein the size of the preset number is 500 or 1000 and so on, the disclosed embodiment is not limited to the specific. The preset number of male front human face Image as the site to obtain the position of the follow-up to the male head and shoulder a priori model training data stored in the Image training in a library.

2nd step, according to the user for each site of the male human face picture of front shoulder prospect calibration result, generates a piece of male human face picture of front shoulder of the calibration Image.

With step 201 a 2nd step in the same token, where does not repeat.

3rd step, for each piece of male front human face Image face external contour characteristic point positioning, the positioning result obtained.

With step 201 a 3rd step in the same token, where does not repeat.

4th step, according to the external outline of the face of the positioning result, each shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image.

With step 201 a in the same 4th step, here does not repeat.

5th step, for multiple preset size in the Image of the same position pixel point, according to the prospects for shoulder calibration result, calculating the point position where the head shoulder and the prospect of a priori probability.

With step 201 a 5th step in the same token, where does not repeat.

It should be explained that, whatever the against the step 201 a or step 201 b for example, the position of each pixel point in the head shoulder and the prospect of a priori probability may be made by the following formula (1) to obtain, according to the prospects for shoulder calibration result, use the following formula, calculating the point position where the head shoulder and the prospect of a priori probability:

Wherein Refer to pixel point (x_i , Y_o ) Present at the head shoulder and the prospect of a priori probability, Refer to section j of images pixel point (x_i , Y_o ) At the calibration result, J site of said pixel in the Image point (x_i , Y_o ) Calibrated at [...] prospect, J site of said pixel in the Image point (x_i , Y_o ) At the calibration into background, N the front side of the indicating the number of the face Image. For the purposes of female front human face Image, in the calculation of each pixel point position where the head and shoulder of the a priori probability after the prospect, to obtain the position of the female head and shoulder a priori model; for the purposes of male front human face Image, in the calculation of each pixel point position where the head and shoulder of the a priori probability after the prospect, get the male head shoulder position a priori model.

In step 202 in, the treatment of the human face in split Image gender identification, gender identification result obtained; according to the results of the gender identification, determining corresponding to the a priori probability of the segmented Image.

In the disclosed embodiment, because the male face Image and female face characteristic of the picture for the difference, corresponding to the different position of the head and shoulder of the a priori model, in the case before the Image segmentation, also need to treat in split Image of the human face gender identification. In the treatment of the human face in the split Image when the gender identification, may take the existing human face recognition algorithm to realize, in this does not then repeat. Because of the above-mentioned step 201 in the position of the female head and shoulder has been a priori model and the position of the male head and shoulder a priori model, therefore in determining to be divided Image of human face in after gender, can be directly determined to be divided Image corresponding to the position of the shoulder of a priori model, also get to be divided Image corresponding to the priori probability.

In step 203 in, to be in split Image for each of a plurality of pixel points, judging the pixel point is greater than the a priori probability of the shoulder prospect probability threshold; if the pixel point of the a priori probability is greater than shoulder prospects probability threshold, then the pixel point as prospect sample pixel point; judging the pixel point of the a priori probability is less than the background probability threshold; if the pixel point of the a priori probability is less than the background probability threshold, then the pixel point as background sample pixel point.

Wherein the head shoulder and the prospect probability threshold specific can be 0.7 or 0.8 and so on, the disclosed embodiment is not limited to the specific. Background probability threshold specific can be 0.2 or 0.3 and so on, the disclosed embodiment of this same are not specifically limited.

In the disclosed embodiment, the selected sample pixel point before, treatment can be first split Image of the human face external contour characteristic point positioning, and according to the positioning result to be divided Image is resized, it with the 400 * 400 of the same size and the size of the template. In this way, to be in split Image of each pixel point are with the 400 * 400 in the template with the pixel corresponding to the position of the point. Therefore 400 * 400 template in the position corresponding to the position where the pixel points on the head shoulder and the prospect of a priori probability, it is to be in split Image with a pixel point corresponding to the position of the a-priori probability.

In order to head and shoulders prospect the probability threshold is 0.8, background probability threshold is 0.2 for example, is for a pixel point, if the pixel point of the a priori probability is greater than 0.8, the pixel point will be determined as the prospect sample pixel point; if the pixel point of the a priori probability is less than 0.2, then the pixel point as background sample pixel point. For the priori probability is in the 0.2 to 0.8 before pixel point is not of any processing.

In step 204 in, according to the prospects for shoulder sample pixel point feature vector and the color of the background sample pixel point of the color feature vector, calculating the head shoulder and the prospect of color likelihood probability and 1st 2nd background of color likelihood probability.

In the disclosed embodiment, the prospects for shoulder sample pixel point feature vector and the color of the background sample pixel point of the color feature vector, may take the existing color vector calculation in a manner, not here to repeat. In the calculation of the 1st and 2nd color color likelihood probability likelihood probability, i.e. before the color likelihood model is obtained, the present invention embodiment also need according to the color of the correlation, the correct shoulder prospect sample picture element spot and background sample pixel point cluster. In the clustering, the disclosed embodiment the head shoulder and the prospects for the sample pixel point cluster into N_F =5 cluster center; taking into account the complexity of the background, the present invention embodiment of the head shoulder and the prospect sample pixel point cluster into N_B =15 cluster center. Of course, N_F And N_B The size of the outside of the addition to the above value, can also be of other values, the disclosed embodiment of this are not specifically limited.

Wherein the shoulder prospect sample pixel point according to the color of the background sample feature vector and the color of the pixel points in the feature vector, calculating the head shoulder and the prospect of the 1st color likelihood probability and background of the 2nd color likelihood probability, by the following formula (2) and formula (3) to realize.

Wherein Refer to section j clusters of central color feature vector, Refer to 1st color likelihood probability, 2nd indicating color likelihood probability, N_F To refer to the head shoulder and the prospect sample pixel point of the number of the cluster center, N_B To refer to the background sample pixel point of the number of the cluster center; Refer to color feature vector To the cluster center European distance, w_j Refer to section j clusters in the center of the sample pixel point of the total sample pixel point ratio, β refer to different cluster of central color feature vector between the average European distance value, it is defined as:

In step 205 in, according to the a priori probability, 1st and 2nd color color likelihood probability likelihood probability, calculating the head shoulder and the prospect of posterior probability and the 1st 2nd background of the posteriori probability.

In the disclosed embodiment, the position of the shoulder in the obtained a priori model and color likelihood models, according to the Bayesian posterior probability theory, the head shoulder and the prospects in the known sample picture element spot and the color of the background sample pixel point of the [...] vector, the prospects of the jackpots posterior probability and background of posterior probability, can be through the following formula formula (6) and formula (7) to obtain.

Wherein To refer to the head shoulder and the prospect of posterior probability;

Background indicating the posterior probability.

In step 206 in, according to the 1st and 2nd of the posterior probability of the posteriori probability, treatment of dividing Image divided head and shoulder.

In the disclosed embodiment, obtaining the posteriori probability 1st and 2nd after the a posteriori probability, to obtain the final the head shoulder and the segmentation model. The head shoulder and the segmentation model into the Graph cut optimization in the context of the data item, and binding the Image pixel contrast information, adopts the min - cut - max - flow optimization approach of dividing the divided Image, can get the final segmentation result of the head and shoulder.

The disclosed embodiment of the method, according to the face due to the external outline of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, based on the a priori probability of the prospects for the shoulder of the head shoulder and the prospects for automatically selecting the sample picture element spot and background sample pixel point, to reach in the sample pixel point selection process is not needed in the involvement of the user, the user experience is good; in addition, in the Image segmentation based on the a priori probability of the color and the likelihood probability divided, segmentation criteria are more detailed, segmentation accuracy is high.

Figure 3 is the block diagram of an exemplary implementation instantiates according to a kind of Image segmentation device. With reference to Figure 3, the device comprises a priori probability calculation module 301, sample pixel point selecting module 302, color likelihood probability calculation module 303, a posteriori probability calculation module 304, Image segmentation module 305.

Wherein the priori probability calculation module 301, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability; sample pixel point selecting module 302 with the a priori probability calculation module 301 is connected, according to the a priori probability, the pre-set shoulder prospect probability threshold and background probability threshold, to be in split Image selecting the head shoulder and the sample picture element spot and the prospects for the background sample pixel point; color likelihood probability calculation module 303 and sample pixel point selecting module 302 is connected to the, according to the prospects for shoulder sample pixel point feature vector and the color of the background sample pixel point of the color feature vector, calculating the head shoulder and the prospect of color likelihood probability and 1st 2nd background of color likelihood probability; a posteriori probability calculation module 304 with the color likelihood probability calculation module 303 is connected, according to the a priori probability, 1st and 2nd color color likelihood probability likelihood probability, calculating the head shoulder and the prospect of posterior probability and the 1st 2nd background of the posteriori probability; Image segmentation module 305 with the posterior probability calculating module 304 is connected to the, for according to the posterior probability and the 1st 2nd a posteriori probability, treatment of dividing Image divided head and shoulder.

Optionally, the a priori probability calculating module, and is used for selecting a preset number of female human face picture of front site; according to the user for each site of female human face picture of front shoulder prospect calibration result, generates a site of female human face picture of front shoulder calibration Image; a site to each female front human face Image face external contour characteristic point positioning, the positioning result obtained; according to face the exterior outline of the positioning result of feature points, each of the shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image; for one of a plurality of preset size in the Image of the same position pixel point, according to the prospects for shoulder calibration result, calculating the point position where the head shoulder and the prospect of a priori probability.

Optionally, the a priori probability calculating module, and is used for selecting a preset number of male human face picture of front site; according to the user for each site of the male human face picture of front shoulder prospect calibration result, generates a site of male human face picture of front shoulder calibration Image; a site to each male front human face Image face external contour characteristic point positioning, the positioning result obtained; according to face the exterior outline of the positioning result of feature points, each of the shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image; for one of a plurality of preset size in the Image of the same position pixel point, according to the prospects for shoulder calibration result, calculating the point position where the head shoulder and the prospect of a priori probability.

Optionally, the a priori probability calculating module, the application of the following formula, calculating the point position where the head shoulder and the prospect of a priori probability:

Wherein Refer to pixel point (x_i , Y_o ) Present at the head shoulder and the prospect of a priori probability, Refer to section j of images pixel point (x_i , Y_o ) At the calibration result, J site of said pixel in the Image point (x_i , Y_o ) Calibrated at [...] prospect, J site of said pixel in the Image point (x_i , Y_o ) At the calibration into background, N the front side of the indicating the number of the face Image.

Optionally, the sample pixel point selecting module, used for the treatment of the human face in split Image gender identification, gender identification result obtained; according to the results of the gender identification, determining corresponding to the a priori probability of a segmented Image; for each of the to be in split Image a pixel point, judging the pixel point is greater than the a priori probability of the shoulder prospect probability threshold; if the pixel point of the a priori probability is greater than shoulder prospects probability threshold, then the pixel point as prospect sample pixel point; for each of the to be in split Image a pixel point, judging the pixel point of the a priori probability is less than the background probability threshold; if the pixel point of the a priori probability is less than the background probability threshold, then the pixel point as background sample pixel point.

Optionally, color likelihood probability calculating module, the application of the following formula, calculating the head shoulder and the prospect of color likelihood probability and 1st 2nd background of color likelihood probability:

Wherein Refer to section j clusters of central color feature vector, Refer to 1st color likelihood probability, 2nd indicating color likelihood probability, N_F To refer to the head shoulder and the prospect sample pixel point of the number of the cluster center, N_B To refer to the background sample pixel point of the number of the cluster center; Refer to color feature vector To the cluster center European distance, w_j Refer to section j clusters in the center of the sample pixel point of the total sample pixel point ratio, β refer to different cluster of central color feature vector between the average European distance value, it is defined as:

Alternatively, the posterior probability calculating module, the application of the following formula, calculate the posterior probability of the head shoulder and the prospects, the posterior probability of the background, including:

Wherein To refer to the head shoulder and the prospect of posterior probability;

Background indicating the posterior probability.

The present invention provided by the embodiment of the device, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, based on the a priori probability of the prospects for the shoulder of the head shoulder and the prospects for automatically selecting the sample picture element spot and background sample pixel point, to reach in the sample pixel point selection process is not needed in the involvement of the user, the user experience is good; in addition, in the Image segmentation based on the a priori probability of the color and the likelihood probability divided, segmentation criteria are more detailed, segmentation accuracy is high.

On the above-mentioned embodiment of the device, wherein each module implementation of the specific mode of operation has been in the the embodiment of the method carried out in the detailed description, the present invention will not be described in detail in the description.

Figure 4 is the block diagram of an exemplary implementation instantiates according to a kind of used for the table-Image segmentation of the apparatus 400. For example, device 400 may be a mobile telephone, computer, digital broadcast terminal, messaging device, game console, flat panel equipment, medical equipment, exercise equipment, a personal digital assistant and the like.

With reference to Figure 4, the apparatus 400 can include the following one or a plurality of components: processing assembly 402, memory 404, the power supply assembly 406, the multimedia component 408, audio assembly 410, I/O (Input/Output, input/output) interface 412, sensor assembly 414, and communication assembly 416.

The processing component 402 usually control apparatus 400 of the overall operation, such as with the display, telephone call, data communication, camera operation and the associated operation of the recording operation. The processing component 402 can include one or more processor 420 to execute the instruction, in order to complete the above-mentioned method of all or part of the steps. In addition, the processing unit 402 can include one or more module, the easy processing assembly 402 and the interaction between the other components. For example, the processing component 402 may include a multimedia module, so as to facilitate the multimedia component 408 and the processing unit 402 of the interaction between the.

Memory 404 is configured to store various types of data in order to support the apparatus 400 of the operation. These data include for example apparatus 400 operates on any application program or method of instruction, Contact Person data, the telephone directory data, message, picture, video and the like. Memory 404 can be any type of volatile or non-volatile storage device or a combination thereof to achieve, such as SRAM (Static Random Access Memory, static random access memory), EEPROM (Electrically - Erasable Programmable Read - Only Memory, electrically erasable programmable read-only memory), EPROM (Erasable Programmable Read Only Memory, erasable programmable read-only memory), PROM (Programmable Read - Only Memory, programmable read-only memory), ROM (Read - Only Memory, read-only memory), magnetic memory, flash memory, magnetic disk or optical disk.

The power supply assembly 406 for the apparatus 400 of the various components to provide power. The power supply assembly 406 may include a power management system, one or more of power, and the other with the equipment 400 generation, management and distribution of power associated with the assembly.

Multimedia component 408 included in the apparatus 400 and between the users of the provides an output interface screen. In some embodiments, the screen may include a LCD (Liquid Crystal Display, liquid crystal display) and TP (Touch Panel, touch panel). If the screen comprises a TP, screen can be implemented as a touch screen, in order to receive the input signal from the user. TP includes one or a plurality of touch sensors in order to detect the touch, sliding and of the gestures on the touch panel. The touch sensor can not only sensing touch or sliding action of the boundary, but also can detect the touch or the sliding operation related and a duration of the pressure. In some embodiments, the multimedia component 408 includes a front camera and/or rear camera. When the apparatus 400 is in the mode of operation, such as the shooting mode or a video mode, front camera and/or rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or having focus and optical zoom capability.

Audio assembly 410 is configured to output and/or input audio signal. For example, audio assembly 410 includes a MIC (Microphone, microphone), when the device 400 is in the mode of operation, such as the call mode, recording mode and voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 404 or via the communication assembly 416 transmission. In some embodiments, audio assembly 410 also includes a loudspeaker, for outputting the audio signal.

I/O interface 412 for processing assembly 402 and a peripheral interface module providing an interface between, the peripheral interface module can be a keyboard, clicking wheel, such as push-button. These button may include but are not limited to: home page button, a volume button, start button and the locking button.

Sensor assembly 414 includes one or a plurality of sensors, is used for device 400 to provide the various aspects of the state evaluation. For example, sensor assembly 414 can detect device 400 of the open/closed state, the relative positioning of the assembly, such as assembly for the apparatus 400 of the display and a keyboard, sensor assembly 414 can also measure the apparatus 400 or apparatus 400 and a second assembly position change, the user with the device 400 contact the presence or absence of the, apparatus 400 orientation or acceleration/deceleration and the apparatus 400 of the temperature change. Sensor assembly 414 may include a proximity sensor, is configured to in the absence of any physical contact with the vicinity of the detecting the existence of the object. Sensor assembly 414 may also include a photo sensor, such as CMOS (Complementary Metal Oxide Semiconductor, complementary metal oxide) or CCD (Charge - coupled Device, charge coupled device) Image sensor, for use in imaging applications. In some embodiments, the sensor assembly 414 may also include acceleration sensor, gyro sensor, magnetic sensor, pressure sensor or a temperature sensor.

Communication module 416 configured to facilitate the apparatus 400 and other equipment between the wired or wireless mode of communication. Device 400 can access based on communication standard wireless network, such as WiFi, 2 G or 3 G, or a combination thereof. In one exemplary embodiment, communication module 416 via the broadcast channel received from the external broadcast management system of the broadcast signal or broadcast related information. In one exemplary embodiment, communication module 416 also includes the NFC (Near Field Communication, near-field communication) module, in order to promote the short-range communication. For example, in the NFC module can be based on RFID (Radio Frequency Identification, radio frequency identification) technology, IrDA (Infra - red Data Association, infrared data association) technology, UWB (Ultra Wideband, ultra-wideband) technology, BT (Bluetooth, Bluetooth) technology and other technologies to realize.

In an exemplary embodiment, the apparatus 400 can be one or more of ASIC (Application Specific Integrated Circuit, an application specific integrated circuit), DSP (Digital signal Processor, digital signal processor), DSPD (Digital signal Processor Device, digital signal processing equipment), PLD (Programmable Logic Device, programmable logic device), FPGA) (Field Programmable Gate Array, field programmable gate array), controller, microcontroller, microprocessor or other electronic elements to realize the, used for carrying out the above-mentioned method.

In an exemplary embodiment, also provides a includes an instruction of a non-transitory computer readable storage medium, such as includes an instruction memory 404, the above-mentioned order by the apparatus 400 of the processor 420 in order to complete the implementation of the above-mentioned method. For example, a non-transitory computer readable storage medium may be a ROM, RAM (Random Access Memory, random access memory), CD - ROM (Compact Disc Read - Only Memory, CD-ROM), magnetic tape, floppy disk and optical data memory device and so on.

A non-transitory computer readable storage medium, when the storage medium by the instructions in the execution processor of the mobile terminal, the mobile terminal can carry out a kind of Image segmentation method, the method comprises:

According to face the external outline of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability;

According to the a priori probability, the pre-set shoulder prospect probability threshold and background probability threshold, to be in split Image selecting the head shoulder and the sample picture element spot and the prospects for the background sample pixel point;

According to the prospects for shoulder sample pixel point feature vector and the color of the background sample pixel point of the color feature vector, calculating the head shoulder and the prospect of color likelihood probability and 1st 2nd background of color likelihood probability;

According to the a priori probability, 1st and 2nd color color likelihood probability likelihood probability, calculating the head shoulder and the prospect of posterior probability and the 1st 2nd background of the posteriori probability;

According to the 2nd 1st posterior probability and the posteriori probability, treatment of dividing Image divided head and shoulder.

Optionally, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, including:

Selecting a preset number of female human face picture of front site;

According to the user for each site of female human face picture of front shoulder prospect calibration result, generates a piece of female human face picture of front shoulder of the calibration Image;

For each piece of female front human face Image face external contour characteristic point positioning, the positioning result obtained;

According to face the exterior outline of the positioning result of feature points, each of the shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image;

For multiple preset size in the Image of the same position pixel point, according to the prospects for shoulder calibration result, calculating the point position where the head shoulder and the prospect of a priori probability.

Optionally, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, including:

The site selected preset number of male front human face Image;

According to the user for each site of the male human face picture of front shoulder prospect calibration result, generates a piece of male human face picture of front shoulder of the calibration Image;

For each piece of male front human face Image face external contour characteristic point positioning, the positioning result obtained;

According to face the exterior outline of the positioning result of feature points, each of the shoulder to a site calibration Image alignment and size normalization processing, get the multiple preset size Image;

For multiple preset size in the Image of the same position pixel point, according to the prospects for shoulder calibration result, calculating the point position where the head shoulder and the prospect of a priori probability.

Optionally, the prospects for the shoulder according to the calibration results, use the following formula, calculating the point position where the head shoulder and the prospect of a priori probability:

Wherein Refer to pixel point (x_i , Y_o ) Present at the head shoulder and the prospect of a priori probability, Refer to section j of images pixel point (x_i , Y_o ) At the calibration result, J site of said pixel in the Image point (x_i , Y_o ) Calibrated at [...] prospect, J site of said pixel in the Image point (x_i , Y_o ) At the calibration into background, N the front side of the indicating the number of the face Image.

Optionally, according to the a priori probability, the pre-set shoulder prospect probability threshold and background probability threshold, to be in split Image selecting the head shoulder and the sample picture element spot and the prospects for the background sample pixel point, including:

Treatment of the human face in split Image gender identification, gender identification result obtained;

According to the results of the gender identification, determining corresponding to the a priori probability of the segmented Image;

For each of the to be in split Image a pixel point, judging the pixel point is greater than the a priori probability of the shoulder prospect probability threshold; if the pixel point of the a priori probability is greater than shoulder prospects probability threshold, then the pixel point as prospect sample pixel point;

For each of the to be in split Image a pixel point, judging the pixel point of the a priori probability is less than the background probability threshold; if the pixel point of the a priori probability is less than the background probability threshold, then the pixel point as background sample pixel point.

Optionally, according to the prospects for shoulder sample pixel point feature vector and the color of the background sample pixel point of the color feature vector, use the following formula, calculating the head shoulder and the prospect of color likelihood probability and 1st 2nd background of color likelihood probability, including:

Wherein Refer to section j clusters of central color feature vector, Refer to 1st color likelihood probability, 2nd indicating color likelihood probability, N_F To refer to the head shoulder and the prospect sample pixel point of the number of the cluster center, N_B To refer to the background sample pixel point of the number of the cluster center; Refer to color feature vector To the cluster center European distance, w_j Refer to section j clusters in the center of the sample pixel point of the total sample pixel point ratio, β refer to different cluster of central color feature vector between the average European distance value, it is defined as:

Optionally, according to the a priori probability, 1st and 2nd color color likelihood probability likelihood probability, the application of the following formula, calculate the posterior probability of the head shoulder and the prospects, the posterior probability of the background, including:

Wherein To refer to the head shoulder and the prospect of posterior probability;

Background indicating the posterior probability.

The disclosed embodiment provides non-temporary computer readable storage medium, according to the external outline of the face of the positioning result, calculating the preset size Image on each pixel in a point position where the head shoulder and the prospect of a priori probability, based on the a priori probability of the prospects for the shoulder of the head shoulder and the prospects for automatically selecting the sample picture element spot and background sample pixel point, to reach in the sample pixel point selection process is not needed in the involvement of the user, the user experience is good; in addition, in the Image segmentation based on the a priori probability of the color and the likelihood probability of dividing the, segmentation criteria are more detailed, segmentation accuracy is high.

Figure 5 is the block diagram of according to another exemplary instantiates a used for Image segmentation apparatus 500. For example, device 500 may be provided as a server. With reference to Figure 5, apparatus 500 comprises a processing component 522, it further includes the one or more processor, and by the memory 532 on behalf of the memory resources, is for storing the processing assembly 522 of the implementation of the instruction, such as the application program. Memory 532 is stored in the application program may include one or more than one each corresponding to a group of instructions of the module. In addition, the processing unit 522 is configured to execute the instruction, in order to carry out the above-mentioned examples provide Image segmentation method.

The apparatus 500 can also include a power supply assembly 526 is configured for the implementation of the apparatus 500 of the power management, a wired or wireless network interface 550 is configured to the device 500 is connected to the network, and an input-output (I/O) interface 558. The apparatus 500 can operate based on stored in the memory 532 of the operating system, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or similar.

The technicians of this field in the consideration of the specification and practice of the invention disclosed here after, will easily think of the other embodiments disclosed. The application intended to cover any variations of the present disclosure, use or adaptive, these variable-type, use or change in accordance with the adaptability of the invention claims the general principle and comprises a the present invention is not disclosed in this technical field of well known and common sense or conventional technical means. Specification and the execution to be regarded as illustrative, the disclosure of the true scope and spirit of the following claim that by.

It should be understood that, the present invention is not limited to the above description and in the attached have been shown in the precise structure, and can be in the without departing from its scope various modification and change. The scope of the disclosed only by the attached claims to limit.