CN104732509A - Self-adaptation image segmentation method and device - Google Patents

Abstract

The invention discloses a self-adaptation image segmentation method and device. The self-adaptation image segmentation method comprises the following steps of extracting a region of interest (ROI) from an input image; extracting one or more features of the ROI and classifying the ROI on the basis of the extracted features; generating different initial foreground distributions for the ROI of different types; segmenting the ROI on the basis of the generated initial foreground distributions so as to obtain a target object region.

Description

Self-adaptive projection method method and apparatus

Technical field

The present invention relates to the Iamge Segmentation in computer vision field, more particularly, relate to a kind of high precision, self-adapting division method and the equipment that are applicable to medical image.

Background technology

In order to obtain the image of inside of human body object, need from the medical images such as ultrasonoscopy, CT image, be partitioned into inside of human body object border and profile.Here, inside of human body object can be organ, tumour, tissue etc.In reality, inside of human body object is very complicated.Such as, mammary tumor has many types, comprise benign tumour (tumour, fibroma etc.) and malignant tumour, the variety classes even tumour of the one species performance in ultrasonoscopy (as bright in planform and systematicness, size, gray scale dark, with the contrast of surrounding tissue, whether have calcification point etc.) has huge diversity, and Accurate Segmentation goes out pathological structure and has great difficulty.

Medical image segmentation can be divided into interactive and full-automatic two kinds of dividing methods.In full-automatic method, automatically detected the general area (that is, area-of-interest (ROI)) at pathology place by system, be then partitioned into the precise boundary of pathological structure; And in man-machine interactively formula method, there are three kinds of usual ways, the first needs the region of a part of pathological structure of artificial input and a part of background area, and the second points out Seed Points in pathological structure region, and the third is the general area (ROI) going out pathological structure with frames such as rectangle frames.

But there are distinct issues in existing image Segmentation Technology.Specifically, have a lot of parameter in image segmentation algorithm, wherein the initial distribution of prospect or background is very large on the impact of final segmentation result, and the partitioning algorithm based on that fix, unified initial distribution production method is then difficult to the huge ROI of reply difference.Therefore, existing image Segmentation Technology is difficult to obtain good segmentation effect.

Summary of the invention

Therefore, an aspect of of the present present invention is to provide a kind of significant characteristics and smoothness feature by extracting ROI to produce different initial prospect distributions, and carries out the method and apparatus of Iamge Segmentation based on initial prospect distribution.

According to an aspect of the present invention, a kind of self-adaptive projection method method is provided, comprises the following steps: extract area-of-interest (ROI) from input picture; Extract the one or more of features of ROI, and based on the feature extracted, ROI is classified; Different initial prospect distributions is produced for dissimilar ROI; ROI is split, to obtain targeted object region based on the initial prospect distribution produced.

Preferably, the feature of extraction is significant characteristics, smoothness feature or their combination.

Preferably, the significant characteristics of ROI is extracted according to following equation:

$\mathrm{SaliencyScore}=\underset{i=0}{\overset{i=\mathrm{countPixel}}{\mathrm{\Σ}}}\mathrm{arraySaliency}\left[\mathrm{grayValue}\right[i\left]\right]*\exp (-\mathrm{dist}/\mathrm{\δ}),$

Wherein, SaliencyScore represents one dimension significant characteristics vector, i represents i-th pixel in ROI, countPixel represents the sum of the pixel in ROI, arraySaliency [] represents pixel value conspicuousness array, grayValue [i] represents the gray-scale value of i-th pixel, and dist represents the distance of i-th pixel to reference position, and δ represents the weights of distance and 0< δ <1.

Preferably, the significant characteristics of ROI is extracted according to following equation:

SaliencyImage[i]=arraySaliency[grayValue[i]]*exp(-dist/δ)，

Wherein, SaliencyImage represents multidimensional significant characteristics vector, SaliencyImage [i] represents the significant characteristics of i-th pixel, arraySaliency [] represents pixel value conspicuousness array, grayValue [i] represents the gray-scale value of i-th pixel, dist represents the distance of i-th pixel to reference position, and δ represents the weights of distance and 0< δ <1.

Preferably, according to following equation determination pixel value conspicuousness array:

$\begin{array}{c}\mathrm{arraySaliency}\left[\mathrm{grayLevel}\right]=\underset{\mathrm{grayLevel}=\min \mathrm{GrayLevel}}{\overset{\mathrm{grayLevel}<=\max \mathrm{GrayLevel}}{\mathrm{\&Sigma;}}}\mathrm{histogram}\left[\mathrm{grayLevel}\right]*\mathrm{abs}(\mathrm{grayLevel}-\mathrm{grayLevel}1)*,\\ \exp (-\mathrm{dist}/\mathrm{\&delta;})\end{array}$

Wherein, grayLevel represents the pixel value of specific pixel in ROI, maxGrayLevel represents max pixel value in ROI, minGrayLevel represents minimum pixel value in ROI, histogram [] represents histogram array, grayLevel1 represents the pixel value of each pixel in ROI except specific pixel, and dist represents the distance of specific pixel to reference position, and δ represents the weights of distance and 0< δ <1.

Preferably, the step that ROI classifies is comprised: using described one or more of feature as input, use support vector machine, K average (K-Means) algorithm or neural network ROI to be categorized as dominant class ROI or recessive class ROI.

Preferably, the step producing different initial prospect distributions comprises: for dominant class ROI, use K average (K-Means) algorithm to be three classes by the pixel cluster in ROI; A class pixel minimum for average pixel value is defined as prospect, a class maximum for average pixel value is defined as background; Use and belong to the pixel of prospect and belong to the pixel structure gauss hybrid models (GMM) of background; To a class pixel of prospect or background be confirmed as input, by the GMM of structure, such pixel be defined as prospect or background, thus produce the distribution of initial prospect.

Preferably, the step producing different initial prospect distributions comprises: for recessive class ROI, using reference position as center, use fixing rectangular area as prospect, thus produce the distribution of initial prospect.

Preferably, reference position represents the center of gravity of initial targeted object region.

Preferably, determine reference position by following steps: by using large law (OSTU) to split ROI, thus obtain the pixel of two types; Calculate the average pixel value of the pixel of two types; The center of gravity in the region pixel of a type little for average pixel value formed is defined as reference position.

According to a further aspect in the invention, a kind of self-adaptive projection method device is provided, comprises: ROI extraction unit, be configured to extract ROI from input picture; ROI feature extraction and taxon, be configured to the one or more of features extracting ROI, and classify to ROI based on the feature extracted; Initial prospect distribution generation unit, is configured to produce different initial prospect distributions for dissimilar ROI; Image segmentation unit, the initial prospect distribution be configured to based on producing is split ROI, to obtain targeted object region.

Preferably, described one or more of feature is significant characteristics, smoothness feature or their combination.

Preferably, ROI feature extraction and taxon are configured to the significant characteristics extracting ROI according to following equation:

$\mathrm{SaliencyScore}=\underset{i=0}{\overset{i=\mathrm{countPixel}}{\mathrm{\&Sigma;}}}\mathrm{arraySaliency}\left[\mathrm{grayValue}\right[i\left]\right]*\exp (-\mathrm{dist}/\mathrm{\&delta;}),$

Wherein, SaliencyScore represents one dimension significant characteristics vector, i represents i-th pixel in ROI, countPixel represents the sum of the pixel in ROI, arraySaliency [] represents pixel value conspicuousness array, grayValue [i] represents the gray-scale value of i-th pixel, and dist represents the distance of i-th pixel to reference position, and δ represents the weights of distance and 0< δ <1.

Preferably, ROI feature extraction and taxon are configured to the significant characteristics extracting ROI according to following equation:

SaliencyImage[i]=arraySaliency[grayValue[i]]*exp(-dist/δ)，

Wherein, SaliencyImage represents multidimensional significant characteristics vector, SaliencyImage [i] represents the significant characteristics of i-th pixel, arraySaliency [] represents pixel value conspicuousness array, grayValue [i] represents the gray-scale value of i-th pixel, dist represents the distance of i-th pixel to reference position, and δ represents the weights of distance and 0< δ <1.

Preferably, ROI feature extraction and taxon are also configured to according to following equation determination pixel value conspicuousness array:

$\begin{array}{c}\mathrm{arraySaliency}\left[\mathrm{grayLevel}\right]=\underset{\mathrm{grayLevel}=\min \mathrm{GrayLevel}}{\overset{\mathrm{grayLevel}<=\max \mathrm{GrayLevel}}{\mathrm{\&Sigma;}}}\mathrm{histogram}\left[\mathrm{grayLevel}\right]*\mathrm{abs}(\mathrm{grayLevel}-\mathrm{grayLevel}1)*,\\ \exp (-\mathrm{dist}/\mathrm{\&delta;})\end{array}$

Wherein, grayLevel represents the pixel value of specific pixel in ROI, maxGrayLevel represents max pixel value in ROI, minGrayLevel represents minimum pixel value in ROI, histogram [] represents histogram array, grayLevel1 represents the pixel value of each pixel in ROI except specific pixel, and dist represents the distance of specific pixel to reference position, and δ represents the weights of distance and 0< δ <1.

Preferably, ROI feature extraction and taxon are configured to using described one or more kinds of feature as input, use support vector machine, K average (K-Means) algorithm or neural network ROI to be categorized as dominant class ROI or recessive class ROI.

Preferably, initial prospect distribution generation unit is configured to: for dominant class ROI, use K average (K-Means) algorithm to be three classes by the pixel cluster in ROI; A class pixel minimum for average pixel value is defined as prospect, a class maximum for average pixel value is defined as background; Use and belong to the pixel of prospect and belong to the pixel structure gauss hybrid models (GMM) of background; To a class pixel of prospect or background be confirmed as input, by the GMM of structure, such pixel be defined as prospect or background, thus produce the distribution of initial prospect.

Preferably, initial prospect distribution generation unit is configured to: for recessive class ROI, using reference position as center, use fixing rectangular area as prospect, thus produce the distribution of initial prospect.

Preferably, reference position represents the center of gravity of initial targeted object region.

Preferably, initial prospect distribution generation unit is also configured to: by using large law (OSTU) to split ROI, thus obtain the pixel of two types; Calculate the average pixel value of the pixel of two types; The center of gravity in the region pixel of a type little for average pixel value formed is defined as reference position.

According to exemplary embodiment of the present invention, undertaken classifying by the one or more of features extracting ROI and produce the distribution of initial prospect based on classification results, the precision of Iamge Segmentation can be improved.

Accompanying drawing explanation

By the description carried out embodiment below in conjunction with accompanying drawing, these and/or other aspect of the present invention and advantage will become clear and be easier to understand, in the accompanying drawings:

Fig. 1 is the process flow diagram of the self-adaptive projection method method illustrated according to exemplary embodiment of the present invention;

Fig. 2 and Fig. 3 illustrates the dominant class ROI that significant characteristics and smoothness feature by extracting ROI are classified to ROI and obtained and recessive class ROI respectively;

Fig. 4 illustrates the result producing the distribution of initial prospect according to the use K-Means algorithm of exemplary embodiment of the present invention in conjunction with GMM model and split ROI;

Fig. 5 illustrates that fixing rectangular area according to the use of exemplary embodiment of the present invention produces the distribution of initial prospect and the result split ROI;

Fig. 6 is the block diagram of the self-adaptive projection method equipment illustrated according to exemplary embodiment of the present invention.

Embodiment

More fully the present invention is described hereinafter with reference to accompanying drawing, exemplary embodiment of the present invention shown in the drawings.But the present invention can implement in many different forms, and should not be interpreted as being confined to proposed embodiment here.On the contrary, provide these embodiments to make the disclosure will be thoroughly with completely, and scope of the present invention is conveyed to those skilled in the art fully.

Although it should be understood that and term first, second, third, etc. can be used here to describe different elements, assembly, region, layer and/or part, these elements, assembly, region, layer and/or part should by the restrictions of these terms.These terms are only used to an element, assembly, region, layer or part and another element, assembly, region, layer or part to make a distinction.Therefore, when not departing from instruction of the present invention, the first element discussed below, assembly, region, layer or part can be referred to as the second element, assembly, region, layer or part.As used herein, term "and/or" comprises one or more combination in any and all combinations of lising of being correlated with.

Term used herein only in order to describe the object of specific embodiment, and is not intended to limit the present invention.As used herein, unless the context clearly indicates otherwise, otherwise singulative be also intended to comprise plural form.It is also to be understood that, " comprise " when using term in this manual and/or " comprising " time, there is described feature, entirety, step, operation, element and/or assembly in explanation, but does not get rid of existence or additional one or more further feature, entirety, step, operation, element, assembly and/or their group.

Unless otherwise defined, otherwise all terms used herein (comprising technical term and scientific terminology) have the meaning equivalent in meaning usually understood with those skilled in the art.It will also be understood that, unless clearly defined here, otherwise term (term such as defined in general dictionary) should be interpreted as having the meaning that in the environment with association area, their meaning is consistent, and will not explained them with desirable or too formal implication.

Hereinafter, the present invention is explained in detail with reference to the accompanying drawings.

Fig. 1 is the process flow diagram of the self-adaptive projection method method illustrated according to exemplary embodiment of the present invention.

With reference to Fig. 1, in step S101, extract area-of-interest (ROI) from input picture.According to exemplary embodiment of the present invention, input picture can be the medical images such as ultrasonoscopy, CT image, MRI image, and ROI represents the region comprising destination object, and wherein, destination object can be the object of inside of human body, as tumour, organ, tissue etc.According to exemplary embodiment of the present invention, extract ROI by auto-initiation scheme or semi-automatic initialization scheme.Such as, auto-initiation scheme uses automatic testing method to come the approximate location of detected target object or border to extract ROI, and semi-automatic initialization scheme can extract ROI based on user's input.But, the invention is not restricted to method described above, but from image, extract ROI by various art methods, be not described in detail here.

In step s 102, extract the one of ROI or more kinds of feature, and based on the feature extracted, ROI is classified.According to exemplary embodiment of the present invention, the feature of extraction can be significant characteristics, their combination of smoothness characteristic sum.But the present invention is not limited thereto, the feature of extraction also can comprise other various features relevant to ROI.Below, be characterized as example with significant characteristics and smoothness to be described.According to exemplary embodiment of the present invention, extract the significant characteristics of ROI by equation (1) or equation (2):

$\mathrm{SaliencyScore}=\underset{i=0}{\overset{i=\mathrm{countPixel}}{\mathrm{\&Sigma;}}}\mathrm{arraySaliency}\left[\mathrm{grayValue}\right[i\left]\right]*\exp (-\mathrm{dist}/\mathrm{\&delta;})---\left(1\right)$

SaliencyImage[i]=arraySaliency[grayValue[i]]*exp(-dist/δ) （2），

Wherein, SaliencyScore represents one dimension significant characteristics vector, SaliencyImage represents multidimensional significant characteristics vector, SaliencyImage [i] represents the significant characteristics of i-th pixel, i represents i-th pixel in ROI, countPixel represents the sum of the pixel in ROI, arraySaliency [] represents pixel value conspicuousness array, grayValue [i] represents the gray-scale value of i-th pixel, dist represents the distance of i-th pixel to reference position, δ represents the weights of distance and 0< δ <1.Pixel value conspicuousness array is determined by equation (3):

$\begin{array}{c}\mathrm{arraySaliency}\left[\mathrm{grayLevel}\right]=\underset{\mathrm{grayLevel}=\min \mathrm{GrayLevel}}{\overset{\mathrm{grayLevel}<=\max \mathrm{GrayLevel}}{\mathrm{\&Sigma;}}}\mathrm{histogram}\left[\mathrm{grayLevel}\right]*\mathrm{abs}(\mathrm{grayLevel}-\mathrm{grayLevel}1)*\\ \exp (-\mathrm{dist}/\mathrm{\&delta;})\end{array}---\left(3\right),$

Wherein, grayLevel represents the pixel value of specific pixel in ROI, maxGrayLevel represents max pixel value in ROI, minGrayLevel represents minimum pixel value in ROI, histogram [] represents histogram array, grayLevel1 represents the pixel value of each pixel in ROI except specific pixel, and dist represents the distance of specific pixel to reference position, and δ represents the weights of distance and 0< δ <1.Here, grayLevel represents the pixel value of current pixel in ROI, grayLevel1 represents pixel value larger than the pixel value of current pixel in each pixel in ROI except current pixel, during concrete calculating, the absolute value of the difference of pixel value larger than the pixel value of current pixel in the pixel value asking for current pixel respectively and other pixels, as abs (grayLevel-grayLevel1), then abs (grayLevel-grayLevel1) is multiplied by corresponding weight exp (-dist/ δ), then product summation is as the value of arraySaliency [grayLevel].

At equation (1) in equation (3), reference position represents the center of gravity of targeted object region initial in ROI.According to exemplary embodiment of the present invention, determine reference position in the roi by operating as follows.First, by using large law (OSTU) to split ROI, thus obtain the pixel of two types.Then, the average pixel value of the pixel of two types obtained is calculated.Finally, the center of gravity in the region pixel of a type little for average pixel value formed is defined as reference position.

On the other hand, according to exemplary embodiment of the present invention, extract the smoothness feature (Smoothness) of ROI by equation (4):

R=1-1/(1+varianceValue)，

$\begin{array}{c}\mathrm{varianceValue}=(\underset{i=.}{\overset{i<256}{\mathrm{\&Sigma;}}}(i-\mathrm{averageValue})*(i-\mathrm{averageValue})*\\ \left(\mathrm{histogram}\right[i]/\mathrm{pixelcount}))\end{array},$

Wherein, R represents smoothness feature (Smoothness), and i represents and is more than or equal to 0 natural number being less than 256, and averageValue represents the mean value of pixel value in ROI, histogram [] represents histogram array, and pixelCount represents the sum of the pixel in ROI.

After the one or more of features (such as, but not limited to, significant characteristics, their combination of smoothness characteristic sum) extracting ROI, based on the one or more of features extracted, ROI can be classified.Specifically, can, using the one or more of features of extraction as input, support vector machine (SVM) be used ROI to be categorized as dominant class or recessive class.Selectively, can, using the one or more of features of extraction as input, K average (K-Means) algorithm or neural network (such as, but not limited to BP neural network) be used to be dominant class and recessive class by the pixel classifications in ROI.Here, when using significant characteristics as input, preferably use one dimension significant characteristics vector SaliencyScore.But, also can use multidimensional significant characteristics vector SaliencyImage [i].Fig. 2 and Fig. 3 illustrates the dominant class ROI that significant characteristics and smoothness feature by extracting ROI are classified to ROI and obtained and recessive class ROI respectively.As shown in Figure 2, in dominant class ROI, comprise and around have obvious background and contrast the darker destination object of ambient background gray scale (such as, tumour) region, at this moment, human eye can contrast ambient background and identify significantly dark-coloured destination object existence from ROI.But, for part include the destination object of calcification point ROI, comprise the ROI that gray scale is the destination object of light tone, the ROI only comprising destination object inside, the destination object comprised and background contrasts ROI etc. low especially, human eye is difficult to Direct Recognition and goes out destination object, and therefore this ROI is called as recessive class ROI.

Especially, before or after step S102, can Image semantic classification be carried out, such as, the process such as filtering and noise reduction be carried out to input picture or ROI.

In step s 103, different initial prospect distributions is produced for dissimilar ROI.Specifically, for dominant class ROI, first K-Means algorithm is used to be three classes by the pixel cluster in ROI.Then, a class pixel of minimum for average pixel value (that is, low gray scale) is defined as prospect, a class of maximum for average pixel value (that is, low gray scale) is defined as background, remain a class pixel and be temporarily defined as unknown class.Next, use belongs to the pixel of prospect and belongs to the pixel structure gauss hybrid models (GMM) of background.Finally, using the input of the pixel of unknown class as the GMM of structure, by the GMM of structure, the pixel of unknown class is defined as prospect or background, thus produces the distribution of initial prospect.On the other hand, for recessive class ROI, can, using reference position as center, use fixing rectangular area as prospect, thus produce the distribution of initial prospect.Here, reference position is determined in the roi by aforesaid operations.Then, fixing rectangular area is determined with reference position to 0.5-0.9 times of (such as, 0.7 times) size of the distance on all directions border of ROI.But, the present invention is not limited thereto, the ratio suitable according to other can determine the size of fixing rectangular area.Selectively, can also computing reference position to the distance on all directions border of ROI, and by outside for each border of ROI extended reference position to the distance on all directions border 0.05-0.3 doubly (such as, 0.1 times), thus generation initial background distributes.Like this, by extracting the one or more of features of ROI (such as, but be not limited to, significant characteristics, smoothness feature or their combination) carry out classifying and produce the distribution of initial prospect based on classification results, can for accurately splitting the basis that ROI provides necessary.

In step S104, ROI can be split, to obtain destination object (such as, tumour, organ, tissue etc.) region based on the initial prospect distribution produced.Such as, can split ROI by using Level Set dividing method based on the initial prospect distribution produced.Selectively, also can split ROI by using Graph Cut algorithm or Grab Cut algorithm based on the initial prospect distribution produced and initial background distribution.In addition, also can based on the initial prospect distribution produced by using region grow(region growing) algorithm, snake algorithm etc. split ROI.Various dividing method described herein easily can be realized according to prior art by those skilled in the art, therefore omits it and describes in detail.

Fig. 4 illustrates the result producing the distribution of initial prospect according to the use K-Means algorithm of exemplary embodiment of the present invention in conjunction with GMM model and split ROI, and Fig. 5 illustrates and fixes according to the use of exemplary embodiment of the present invention the result that rectangular area produces the distribution of initial prospect and split ROI.

Fig. 6 is the block diagram of the self-adaptive projection method equipment illustrated according to exemplary embodiment of the present invention.

With reference to Fig. 6, self-adaptive projection method equipment 600 comprises ROI extraction unit 610, ROI feature extraction and taxon 620, initial prospect distribution generation unit 630 and image segmentation unit 640.

ROI extraction unit 610 can extract ROI from input picture.As mentioned above, input picture can be the medical images such as ultrasonoscopy, CT image, MRI image, and ROI represents the region comprising destination object, and wherein, destination object can be the object of inside of human body, as tumour, organ, tissue etc.ROI extraction unit 610 extracts ROI by various art methods from image.

ROI feature extraction and taxon 620 can extract the one or more of features of ROI (such as, but be not limited to, significant characteristics, smoothness feature or their combination, or other various features relevant to ROI), and based on the one or more of features extracted, ROI is classified.As mentioned above, ROI feature extraction and taxon 620 extract the significant characteristics of ROI by equation (1) or equation (2), and extract the smoothness feature (Smoothness) of ROI by equation (4).When extracting the significant characteristics of ROI, ROI feature extraction and taxon 620 determine pixel value conspicuousness array by equation (3), and can determine the reference position in ROI.In addition, ROI feature extraction and taxon 620 can based on the one or more of features extracted (such as, but be not limited to, significant characteristics, their combination of smoothness characteristic sum), use SVM, K-Means algorithm or neural network ROI to be categorized as dominant class ROI or recessive class ROI.

Initial prospect distribution generation unit 630 can produce different initial prospect distributions for dissimilar ROI.As mentioned above, for dominant class ROI, initial prospect distribution generation unit 630 can use K-Means algorithm to produce the distribution of initial prospect in conjunction with GMM model, and for recessive class ROI, initial prospect distribution generation unit 630 can use fixing rectangular area to produce the distribution of initial prospect.Selectively, initial prospect distribution generation unit 630 also can produce initial background distribution.

Image segmentation unit 640 can be split ROI based on the initial prospect distribution produced, to obtain targeted object region.As mentioned above, image segmentation unit 640 can use Level Set dividing method, Graph Cut algorithm, Grab Cut algorithm, region grow(region growing) dividing method of the various prior art such as algorithm, snake algorithm splits ROI, thus obtains targeted object region.

According to exemplary embodiment of the present invention, by extracting the one or more of features of ROI ((such as, but be not limited to, significant characteristics, smoothness feature or their combination, or other various features relevant to ROI) carry out classifying and produce the distribution of initial prospect based on classification results, the precision of Iamge Segmentation can be significantly improved.Specifically, all increase significantly according to the Sensitivity rate of the self-adaptive projection method method and apparatus of exemplary embodiment of the present invention, Jacobi's value (jaccard value) and doctor's receptance.

The above-mentioned self-adaptive projection method method according to exemplary embodiment of the present invention can be implemented as software or computer code or their combination.In addition, software or computer code also can be stored in non-transitory recording medium (ROM (read-only memory) (ROM), random-access memory (ram), compact disk (CD)-ROM, tape, floppy disk, optical data storage device and carrier wave (such as being transmitted by the data of internet)) in or by the computer code of web download, wherein, described computer code is initially stored in remote logging medium, computer readable recording medium storing program for performing, or non-transitory machine readable media also will be stored on local recording medium, thus method described herein can use multi-purpose computer, digital machine or application specific processor are to store such software on the recording medium, computer code, software module, software object, instruction, application program, applet, app etc. implement, or be implemented in programmable hardware or specialized hardware (such as ASIC or FPGA).As understood in the art: computing machine, processor, microprocessor controller or programmable hardware comprise volatibility and/or nonvolatile memory and memory assembly (such as RAM, ROM, flash memory etc.), wherein, described storer and memory component can store or receive software or computer code, wherein, described software or computer code will be will be implemented disposal route described herein by computing machine, processor or hardware access when performing.In addition, will recognize: when the code for being implemented on the process shown in this accessed by multi-purpose computer, multi-purpose computer is changed into the special purpose computer for being executed in the process shown in this by the execution of described code.In addition, program can pass through any medium (such as, by wire/wireless connect send signal of communication and equivalent) electronically transmitted.Described program and computer readable recording medium storing program for performing also can be distributed in the computer system of networking, thus store and computer readable code executed with the form of distribution.

Although shown and described some embodiments, it should be appreciated by those skilled in the art that without departing from the principles and spirit of the present invention, can modify to these embodiments, scope of the present invention is by claim and equivalents thereof.

Claims (22)

1. a self-adaptive projection method method, comprises the following steps:

Region of interest ROI is extracted from input picture;

Extract the one or more of features of ROI, and based on the feature extracted, ROI is classified;

Different initial prospect distributions is produced for dissimilar ROI;

ROI is split, to obtain targeted object region based on the initial prospect distribution produced.

2. self-adaptive projection method method according to claim 1, wherein, the feature of extraction is significant characteristics, smoothness feature or their combination.

3. self-adaptive projection method method according to claim 2, wherein, extract the significant characteristics of ROI according to following equation:

$\mathrm{SaliencyScore}=\underset{i=0}{\overset{i=\mathrm{countPixel}}{\mathrm{\&Sigma;}}}\mathrm{arraySaliency}\left[\mathrm{grayValue}\right[i\left]\right]*\exp (-\mathrm{dist}/\mathrm{\&delta;}),$

Wherein, SaliencyScore represents one dimension significant characteristics vector, i represents i-th pixel in ROI, countPixel represents the sum of the pixel in ROI, arraySaliency [] represents pixel value conspicuousness array, grayValue [i] represents the gray-scale value of i-th pixel, and dist represents the distance of i-th pixel to reference position, and δ represents the weights of distance and 0< δ <1.

4. self-adaptive projection method method according to claim 2, wherein, extract the significant characteristics of ROI according to following equation:

SaliencyImage[i]=arraySaliency[grayValue[i]]*exp(-dist/δ)，

Wherein, SaliencyImage represents multidimensional significant characteristics vector, SaliencyImage [i] represents the significant characteristics of i-th pixel, arraySaliency [] represents pixel value conspicuousness array, grayValue [i] represents the gray-scale value of i-th pixel, dist represents the distance of i-th pixel to reference position, and δ represents the weights of distance and 0< δ <1.

5. self-adaptive projection method method according to claim 3, wherein, according to following equation determination pixel value conspicuousness array:

$\begin{array}{c}\mathrm{arraySaliency}\left[\mathrm{grayLevel}\right]=\underset{\mathrm{grayLevel}=\min \mathrm{GrayLevel}}{\overset{\mathrm{grayLevel}<=\max \mathrm{GrayLevel}}{\mathrm{\&Sigma;}}}\mathrm{histogram}\left[\mathrm{grayLevel}\right]*\mathrm{abs}(\mathrm{grayLevel}-\mathrm{grayLevel}1)*,\\ \exp (-\mathrm{dist}/\mathrm{\&delta;})\end{array}$

Wherein, grayLevel represents the pixel value of specific pixel in ROI, maxGrayLevel represents max pixel value in ROI, minGrayLevel represents minimum pixel value in ROI, histogram [] represents histogram array, grayLevel1 represents the pixel value of each pixel in ROI except specific pixel, and dist represents the distance of specific pixel to reference position, and δ represents the weights of distance and 0< δ <1.

6. self-adaptive projection method method according to claim 4, wherein, according to following equation determination pixel value conspicuousness array:

$\begin{array}{c}\mathrm{arraySaliency}\left[\mathrm{grayLevel}\right]=\underset{\mathrm{grayLevel}=\min \mathrm{GrayLevel}}{\overset{\mathrm{grayLevel}<=\max \mathrm{GrayLevel}}{\mathrm{\&Sigma;}}}\mathrm{histogram}\left[\mathrm{grayLevel}\right]*\mathrm{abs}(\mathrm{grayLevel}-\mathrm{grayLevel}1)*,\\ \exp (-\mathrm{dist}/\mathrm{\&delta;})\end{array}$

Wherein, grayLevel represents the pixel value of specific pixel in ROI, maxGrayLevel represents max pixel value in ROI, minGrayLevel represents minimum pixel value in ROI, histogram [] represents histogram array, grayLevel1 represents the pixel value of each pixel in ROI except specific pixel, and dist represents the distance of specific pixel to reference position, and δ represents the weights of distance and 0< δ <1.

7. self-adaptive projection method method according to claim 1, wherein, the step that ROI classifies is comprised: using described one or more of feature as input, use support vector machine, K average K-Means algorithm or neural network ROI to be categorized as dominant class ROI or recessive class ROI.

8. self-adaptive projection method method according to claim 7, wherein, the step producing different initial prospect distributions comprises:

For dominant class ROI, K average K-Means algorithm is used to be three classes by the pixel cluster in ROI;

A class pixel minimum for average pixel value is defined as prospect, a class maximum for average pixel value is defined as background;

Use and belong to the pixel of prospect and belong to the pixel structure gauss hybrid models GMM of background;

To a class pixel of prospect or background be confirmed as input, by the GMM of structure, such pixel be defined as prospect or background, thus produce the distribution of initial prospect.

9. self-adaptive projection method method according to claim 7, wherein, the step producing different initial prospect distributions comprises:

For recessive class ROI, using reference position as center, use fixing rectangular area as prospect, thus produce the distribution of initial prospect.

10. according to the self-adaptive projection method method in claim 3,4,5,6 and 9 described in any one claim, wherein, reference position represents the center of gravity of initial targeted object region.

11. self-adaptive projection method methods according to claim 10, wherein, determine reference position by following steps:

By using large law OSTU to split ROI, thus obtain the pixel of two types;

Calculate the average pixel value of the pixel of two types;

The center of gravity in the region pixel of a type little for average pixel value formed is defined as reference position.

12. 1 kinds of self-adaptive projection method devices, comprising:

ROI extraction unit, is configured to extract ROI from input picture;

ROI feature extraction and taxon, be configured to the one or more of features extracting ROI, and classify to ROI based on the feature extracted;

Initial prospect distribution generation unit, is configured to produce different initial prospect distributions for dissimilar ROI;

Image segmentation unit, the initial prospect distribution be configured to based on producing is split ROI, to obtain targeted object region.

13. self-adaptive projection method devices according to claim 12, wherein, described one or more of feature is significant characteristics, smoothness feature or their combination.

14. self-adaptive projection method devices according to claim 13, wherein, ROI feature extraction and taxon are configured to the significant characteristics extracting ROI according to following equation:

$\mathrm{SaliencyScore}=\underset{i=0}{\overset{i=\mathrm{countPixel}}{\mathrm{\&Sigma;}}}\mathrm{arraySaliency}\left[\mathrm{grayValue}\right[i\left]\right]*\exp (-\mathrm{dist}/\mathrm{\&delta;}),$

Wherein, SaliencyScore represents one dimension significant characteristics vector, i represents i-th pixel in ROI, countPixel represents the sum of the pixel in ROI, arraySaliency [] represents pixel value conspicuousness array, grayValue [i] represents the gray-scale value of i-th pixel, and dist represents the distance of i-th pixel to reference position, and δ represents the weights of distance and 0< δ <1.

15. self-adaptive projection method devices according to claim 13, wherein, ROI feature extraction and taxon are configured to the significant characteristics extracting ROI according to following equation:

SaliencyImage[i]=arraySaliency[grayValue[i]]*exp(-dist/δ)，

Wherein, SaliencyImage represents multidimensional significant characteristics vector, SaliencyImage [i] represents the significant characteristics of i-th pixel, arraySaliency [] represents pixel value conspicuousness array, grayValue [i] represents the gray-scale value of i-th pixel, dist represents the distance of i-th pixel to reference position, and δ represents the weights of distance and 0< δ <1.

16. self-adaptive projection method devices according to claim 14, wherein, ROI feature extraction and taxon are also configured to according to following equation determination pixel value conspicuousness array:

$\begin{array}{c}\mathrm{arraySaliency}\left[\mathrm{grayLevel}\right]=\underset{\mathrm{grayLevel}=\min \mathrm{GrayLevel}}{\overset{\mathrm{grayLevel}<=\max \mathrm{GrayLevel}}{\mathrm{\&Sigma;}}}\mathrm{histogram}\left[\mathrm{grayLevel}\right]*\mathrm{abs}(\mathrm{grayLevel}-\mathrm{grayLevel}1)*,\\ \exp (-\mathrm{dist}/\mathrm{\&delta;})\end{array}$

Wherein, grayLevel represents the pixel value of specific pixel in ROI, maxGrayLevel represents max pixel value in ROI, minGrayLevel represents minimum pixel value in ROI, histogram [] represents histogram array, grayLevel1 represents the pixel value of each pixel in ROI except specific pixel, and dist represents the distance of specific pixel to reference position, and δ represents the weights of distance and 0< δ <1.

17. self-adaptive projection method devices according to claim 15, wherein, ROI feature extraction and taxon are also configured to according to following equation determination pixel value conspicuousness array:

$\begin{array}{c}\mathrm{arraySaliency}\left[\mathrm{grayLevel}\right]=\underset{\mathrm{grayLevel}=\min \mathrm{GrayLevel}}{\overset{\mathrm{grayLevel}<=\max \mathrm{GrayLevel}}{\mathrm{\&Sigma;}}}\mathrm{histogram}\left[\mathrm{grayLevel}\right]*\mathrm{abs}(\mathrm{grayLevel}-\mathrm{grayLevel}1)*,\\ \exp (-\mathrm{dist}/\mathrm{\&delta;})\end{array}$

Wherein, grayLevel represents the pixel value of specific pixel in ROI, maxGrayLevel represents max pixel value in ROI, minGrayLevel represents minimum pixel value in ROI, histogram [] represents histogram array, grayLevel1 represents the pixel value of each pixel in ROI except specific pixel, and dist represents the distance of specific pixel to reference position, and δ represents the weights of distance and 0< δ <1.

18. self-adaptive projection method devices according to claim 12, wherein, ROI feature extraction and taxon are configured to using described one or more kinds of feature as input, use support vector machine, K average K-Means algorithm or neural network ROI to be categorized as dominant class ROI or recessive class ROI.

19. self-adaptive projection method devices according to claim 18, wherein, initial prospect distribution generation unit is configured to: for dominant class ROI, use K average K-Means algorithm to be three classes by the pixel cluster in ROI; A class pixel minimum for average pixel value is defined as prospect, a class maximum for average pixel value is defined as background; Use and belong to the pixel of prospect and belong to the pixel structure gauss hybrid models GMM of background; To a class pixel of prospect or background be confirmed as input, by the GMM of structure, such pixel be defined as prospect or background, thus produce the distribution of initial prospect.

20. self-adaptive projection method devices according to claim 18, wherein, initial prospect distribution generation unit is configured to: for recessive class ROI, using reference position as center, use fixing rectangular area as prospect, thus produce the distribution of initial prospect.

21. according to the self-adaptive projection method device in claim 14,15,16,17 and 20 described in any one claim, and wherein, reference position represents the center of gravity of initial targeted object region.

22. self-adaptive projection method devices according to claim 21, wherein, initial prospect distribution generation unit is also configured to: by using large law OSTU to split ROI, thus obtain the pixel of two types; Calculate the average pixel value of the pixel of two types; The center of gravity in the region pixel of a type little for average pixel value formed is defined as reference position.