CN1700238A - Method for Segmenting Human Skin Regions in Color Digital Images and Videos - Google Patents

Abstract

The invention relates to a method for dividing human body skin area at colorful digital image and video in the field of digital image processing technique. It comprises three main algorisms: incremental Gaussian mixing module algorism, area incremental algorism based on brick, and boundary field potential algorism. The invention starts form 'ordinary skin color module'; skin color catalog of specific image which is obtained from area incremental cycle corrects skin color module and obtains 'special skin color module'; at last it establishes 'boundary field potential' to compensate the gap of skin area at skin area image which is extracted from special skin color module.

Description

In color digital image and video, be partitioned into the method for human body skin area

Technical field

The invention belongs to the digital image processing techniques field, be specifically related to the analysis and the filter method of a kind of medium (image and video) content, further then relate to a kind of method that human body skin area is cut apart that in digital picture and video, realizes.This method can be used for the detection and the filtration of unhealthy on the internet (yellow) media content, stops spreading of unhealthy content, the physical and mental health of protecting young people.

Technical background

Today, along with the raising of Internet development and living standards of the people, the information on the network is more and more abundanter, and the approach that the network user obtains various information is more and more convenient, and speed is also more and more faster.Yet, when the internet brings spiritual wealth for the numerous network users, some lawless persons take advantage of the occasion to have injected waste matter to network for all purposes---and increasing porn site is at every moment threatening numerous netizens, especially teenager netizen's physical and mental health.According to center (XXXcounter) report of pornographic statistics, the whole world has 230,000 porn sites approximately at present, and all increase in the speed with 200-300 every day.For this reason, network is eradicated pornography to struggle against and is fired, and the internet media content filtering system also arises at the historic moment.

More existing network filtering system mostly are based on the filtration of URL address, this technology just directly masks website in the porn site URL address database (being commonly called as " blacklist ") that is set in the Virtual network operator fire wall to the network user simply, rarely has directly the commercial system at the network media (image and video) information filtering both at home and abroad on the market.Though the shield technology based on URL is simply efficient, it has critical limitations: the blacklist because Virtual network operator can't upgrade in time, and this will cause much newly-increased porn site to become fish that has escape the net; Simultaneously, the content of not all webpage all is unsound under some domain name, and this will cause some normal content to be shielded by an innocent person again.Filtration based on media content does not but have above limitation, this technology is directly to carry out the real time content analysis at the some media object on the network (as piece image, one section video), download to the moment that client browses the network user media content is made whether healthy judgement, so determine whether this media object allows to download to client---content-based filtering technique must be the development trend of internet filtering system.Yet, because content-based filtering system need be carried out intellectual analysis to media object, analysis for pornographic image (video) mainly is again to rely on the human body skin area cutting techniques, unfortunately, up to the present, human body skin area cutting techniques itself be exactly one can't fine solution a difficult problem

In color digital image (frame of video), the color of human body skin is very violent owing to the influence that is subjected to two principal elements changes, and these two factors are: (1) intrinsic colour of skin.Different ethnic groups, as white people, yellow etc., and different physical qualifications, as different sexes, age etc., all can have the diverse intrinsic colour of skin; (2) light source condition.Under violent illumination, some can present extremely unsaturation of color to smooth bark skin zone, even bleach fully, and skin area backlight then can deepening; Skin equally also can reflect ambient light, just presents light brown because of contiguous brown floor reflective as the wall that whitewashes white.It is extremely insensitive that the skin color that human vision system causes for these factors changes, and this phenomenon can be explained by Land " the constant phenomenon of color (Color Constancy Phenomenon) " [13]; Yet digital device but can be caught these variations of skin color accurately, objectively in imaging process, and this " objectivity " directly caused " general complexion model " can't comprise all colour of skin kinds, i.e. the embarrassment of " universal model is not general ".Certainly, comprising all skin colors is not a difficult matter, and still the prerequisite here is, complexion model can not also extract the background color except that the colour of skin simultaneously, otherwise complexion model will be without any meaning.

Our find through a large amount of observations, and the limitation of tradition " general complexion model " or " general colour model " ([9,11,15,20]) can be summarized by two pairs of fatal contradictions: (1) versatility and recall ratio.The performance of " general complexion model " normally finds the balance an of the best between it holds the ability of general colour information and special colour of skin information.If too emphasize versatility, distribute to all too dispersions and faint of energy of the various colours of skin in the model, most sample can both obtain certain degree of confidence during detection, but enough high, does not reach colour of skin threshold value, thereby causes low recall ratio; (2) integrality and accuracy rate.The skin area that utilizes " general complexion model " to detect is all very coarse mostly, out of true, because in specific image, because illumination and shade, background color more approaches general skin color than the real skin color in this image under a lot of situations, therefore, background often can obtain the degree of confidence higher than real skin zone, if loosen threshold value and want skin shadow region also extracted because emphasizing the integrality that skin area extracts this moment, the more background area of that consequence is also corresponding to be extracted out, thereby cause low accuracy rate [5,6,10,16,21,22].

Above problem of analysis explanation, exactly " general complexion model " in other words " general colour model (Generic SkinModel) " be impossible all effective to all images, it can only be got a balance between general and special and make maximizing performance.Make complexion model all effective to all images, unique method is set up one " special-purpose complexion model " for each width of cloth specific image exactly.The present invention excites thus, adopt the thought of " from generally to special ", designed a kind of techniqueflow (comprising three main algorithm) of novelty: from " general colour model ", in testing process, from specific testing image, obtain specific colour of skin sample in real time, utilize this specific sample collection online, set up at last " special-purpose complexion model " the correction that circulates of original universal model.

List of references

[1]Adams，R.and?Bischof，L.Seeded?Region?Growing，IEEE?Trans.on?Pattern?Analysis?and?MachineIntelligence，1994，vol.16，pp.641-647.

[2]Canny，J.A?Computational?Approach?to?Edge?Detection，IEEE?Trans.on?Pattern?Analysis?and?MachineIntelligence，1986，vol.8，pp.679-698.

[3]Chang，Y.L.and?Li，X.Adaptive?Image?Region-Growing，IEEE?Trans.on?Image?Processing，1994，vol.3，pp.868-872.

[4]Figueiredo，M.and?Jain，A.K.Unsupervised?Learning?of?Finite?Mixture?Models，IEEE?Trans.on?PatternAnalysis?and?Machine?Intelligence，March?2002，vol.24，no.3，pp.381-396.

[5]Fleck，M.M.，Forsyth，D.A.and?Bregler，C.Finding?Naked?People，In?Proc.of?the?Fourth?European?Confon?Computer?Vision，Berlin，Germany，1996，vol.2，pp.593-602.

[6]Forsyth，D.A.and?Fleck，M.M.Identifying?Nude?Pictures，In?Proc.of?the?3rd?IEEE?Workshop?onApplications?of?Computer?Vision，Dec?2-4，1996，pp.103-108.

[7]Haddon，J.and?Boyce，J.Image?Segmentation?by?Unifying?Region?and?Boundary?Information，IEEETrans?on?Pattern?Analysis?and?Machine?Intelligence，1990，vol.12.

[8]Hall，P.and?Hicks，Y.A?Method?to?Add?Gaussian?Mixture?Models，Tech.Rep.2004-03，Dept.ofComputer?Science，Univ.of?Bath，UK，April?2004.

[9]Jedynak，B.，Zheng，H.，Daoudi，M.and?Barret，D.Maximum?Entropy?Models?for?Skin?Detection，Tech.Rep.XIII，Universite?des?Sciences?et?Technologies?de?Lille，France.

[10]Jeong，C.Y.，Kim，J.S.and?Hong，K.S.Appearance-Based?Nude?Image?Detection，In?Proc.of?the?17 ^th?Int’lConf.on?Pattern?Recognition，Aug?23-26，2004，vol.4，pp.467-470.

[11]Jones，M.J.and?Rehg，J.M.Statistical?Color?Models?with?Application?to?Skin?Detection，In?Proc.of?theComputer?Vision?and?Pattern?Recognition，1999，vol.1，pp.274-280.

[12]Kimmel，R.，Elad，M.，Shaked，D.，Keshet，R.and?Sobel，I.A?Variational?Framework?for?Retinex，HewlettPackard?Tech.Rep.HPL-1999-151，June?1999.

[13]Land，E.H.The?Retinex?Theory?of?Color?Vision，Scientific?American，1977，vol.237.

[14]Pavlidis，T.and?Liow，Y.T.Integrating?Region?Growing?and?Edge?Detection，IEEE?Trans.on?PatternAnalysis?and?Machine?Intelligence，1990，vol.12，pp.225-233.

[15]Phung，S.L.，Bouzerdoum，A.and?Chai，D.Skin?Segmentation?Using?Color?Pixel?Classification：Analysisand?Comparison，IEEE?Trans.on?Pattern?Analysis?and?Machine?Intelligence，January，2005，vol.27，no.1，pp.148-154.

[16]Phung，S.L.，Chai，D.and?Bouzerdoum，A.Adaptive?Skin?Segmentation?in?Color?Images，In?Proc.of?theIEEE?Int’l?Conf.on?Acoustics，Speech，and?Signal?Processing，April?6-10，2003，vol.3，pp.353-356.

[17]Piater，J.H.Mixture?Models?and?Expectation-Maximization，Lecture?at?ENSIMAG，May?2002，updated?onNov?15，2004.

[18]Roberts，S.J.and?Rezek，L.Bayesian?Approaches?to?Gaussian?Mixture?Modeling，IEEE?Trans.on?PatternAnalysis?and?Machine?Intelligence，November?1998，vol.20，no.11.

[19]Shaked，D.and?Keshet，R.Robust?Recursive?Envelope?Operators?for?Fast?Retinex，Hewlett?Packard?Tech.Rep.HPL-2002-74(R.1)，Mar?11，2004.

[20]Zarit，B.D.，Super，B.J?and?Quek，K.H.Comparison?of?Five?Color?Models?in?Skin?Pixel?Classification，InProc.of?the?ICCV’99?Int’l?Workshop?on?Recognition，Analysis，and?Tracking?of?Faces?and?Gestures?inReal-time?System，September，1999，pp.58-63.

[21]Zheng，Q.F.，Zhang，M.J.and?Wang，W.Q.A?Hybrid?Approach?to?Detect?Adult?Web?Images，PCM?2004，Springer-Verlag，Berlin，Heidelberg，2004，pp.609-616.

[22]Zhu，Q.，Wu，C.T.，Cheng，K.T.and?Wu，Y.L.An?Adaptive?Skin?Model?and?Its?Application?toObjectionable?Image?Filtering，In?Proc.of?the?ACM?Int’l?Conf.on?Multimedia(MM’04)，New?York，NY，USA，Oct?10-16，2004.

[23]Zivkovic，Z.and?Heijden，F.Recursive?Unsupervised?Learning?of?Finite?Mixture?Models，IEEE?Trans?onPattern?Analysis?and?Machine?Intelligence，May?2004，vol.26，no.5.

Summary of the invention

The objective of the invention is to propose a kind ofly in color digital image and video, to be partitioned into complete, edge clear, the method for strong semantic human body skin area is arranged.

The method that in color digital image and video, is partitioned into human body skin area that the present invention proposes, be the specific colour of skin sample that utilizes online acquisition to the correction that circulates of original complexion model, its concrete steps are: (1) processed offline (Offline-Process).The a large amount of general colour pixel samples of manual extraction from the various images that contain the different human body skin area of magnanimity are set up " general colour model " with EM (Expectation Maximization) [17] algorithm; (2) pre-service (Pre-Process).From testing image, detect the Canny edge, extract main seed region (Primary Seeded Region) as growth source with " general colour model ", work is done light source compensation (Illumination Compensation) at last based on the region growing at edge; (3) increment is handled (Incremental-Process).Carry out region growing from main seed region based on the edge, obtain skin pixel sample specific in this image, use increment type gauss hybrid models algorithm and carry out online complexion model correction, from this image, extract skin area once more, carry out region growing with the model that newly obtains, online correction, at this moment circulation just can obtain being directed to " the special-purpose complexion model " of this specific image up to this complexion model convergence; (4) aftertreatment (Post-Process).Extract skin area with " special-purpose complexion model ", generate accurate skin area figure, on this figure, set up at last " border potential field (Boundary Potential Field) ", flow in order to the guiding skin area that the lower place of potential energy is filled up because the space in the skin area that shade fold etc. cause.

The skin area figure that extracts by the method for above introduction compared to traditional methods, have two big advantages: accuracy rate is high strong with integrality.It is very low that the former is embodied in the skin area false alarm rate that is partitioned into the skin area cutting techniques among the present invention, and promptly non-skin area (background) flase drop is that the part of skin area is considerably less, and this can guarantee that the skin area that is partitioned into all is real; It is very complete that the latter is embodied in the skin area that is partitioned into the skin area cutting techniques among the present invention, not scattered, discontinuous fritter even pixel, but the cut zone with certain semanteme (skin area that promptly splits is that whole human body, trunk, leg etc. have semantic unit) of bulk, disconnected fritter is then as noise remove.Two characteristics of the present invention just in time satisfy carries out the application demand that high-level semantic is analyzed to the human body skin area that splits, and the classification of yellow image is exactly that it is mainly used.

Description of drawings

Fig. 1 is the flow process frame diagram of algorithm of the present invention.

Fig. 2 is the cutting procedure of a demonstration algorithm flow example of the present invention.

Number in the figure: 1 is the processed offline functional module; 2 is the preprocessing function module; 3 is increment processing capacity module; 4 is the post-processing function module.

Embodiment

Key point of the present invention is three core algorithms and an application framework.Three core algorithms are: (1) increment type gauss hybrid models (Incremental Gaussian Mixture Model) algorithm; (2) based on region growing (the Edge-based Region Growing) algorithm at edge; (3) border potential field (Boundary Potential Field) algorithm.Utilize above three core algorithms designed a kind of in digital picture (video) the cutting techniques framework of human body skin area.

Introduce the present invention is cut apart the flow process framework of human body skin area and formed this framework in digital picture (video) three core algorithms below in detail:

1. the flow process framework of algorithm

This framework can roughly be divided into four main modular: processed offline (Offline-Process), pre-service (Pre-Process), increment are handled (Incremental-Process), aftertreatment (Post-Process).In three core algorithms two, increment type gauss hybrid models and will be used for the increment processing module based on the region growing at edge, and the border potential field will be used for post-processing module.In preceding two modules of this application framework, also will use some present algorithm and technology of comparative maturity.

At this, at first define some hereinafter notions agreements and with the symbol of using.Used in this manual color space is YUV, all mentioned complexion models, comprise " general colour model ", " special-purpose complexion model " and all complexion models with symbol Θ (n) expression, its mathematics prototype all is gauss hybrid models (Gaussian MixtureModels).Θ (n) is illustrated in and is in increment type round-robin complexion model, wherein n 〉=0 in this application framework the n time; This that is to say initial " the general colour model " of Θ (0) expression, and Θ (n) (n＞0) is illustrated in the complexion model that calculates after the n time circulation in the increment processing module, and what obtain at last then is " special-purpose complexion model ".Accompanying drawing 1 is the process flow diagram of this application framework, and label 1-4 represents four above-mentioned main functional modules respectively, supposes input testing image S.

(1) processed offline (Offline-Process): a large amount of general colour pixel samples of manual extraction are as training set from the various images that contain the different human body skin area of magnanimity, with EM (Expectation Maximization) algorithm training " general colour model ", wherein Gauss model quantity is determined by BIC (Bayesian Information Criterion) [18].

(2) pre-service (Pre-Process): from testing image S, detect the Canny edge.Extract the skin pixel zone that reaches certain " scale " with " general colour model ", promptly main seed region (Primary Seeded Region) is as growth source, and the discrete zonule that does not reach certain " scale " will be cast out as noise.Term " scale " is explained intuitively and just is meant that skin pixel reaches a zone that interconnects between certain proportion and skin pixel in the moving window of specifying size herein.Main seed region must be guaranteed the accuracy rate near 100%.After having extracted main seed region, original image S is carried out light source compensation (Illumination Compensation) [12,19].

(3) increment is handled (Incremental-Process): the main seed region that extracts from " general colour model " begins region growing (the growth rule is based on color distortion and edge constraint), can obtain skin pixel sample set specific among this image S; The colour of skin that these new samples that obtain from specific image can be expressed in this image better distributes, adopt increment type mixed Gauss model algorithm, carry out online correction (increment type modeling) with the former complexion model Θ of new samples set pair (0), obtain new complexion model Θ (1), loop ends for the first time; And then with Θ (1) this image is carried out second and take turns the skin area extraction, obtain equally carrying out region growing behind the seed region.Owing to used revised complexion model, screening this time will obtain more complete more accurate area of skin color, and the colour of skin sample that increases newly will reuse increment type mixed Gauss model algorithm and Θ (1) is revised obtain Θ (2), and the rest may be inferred.The area of skin color that is filtered out as Θ (n) is complete to can't expand the time, the increment type mixed Gauss model just can guarantee convergence (| Θ (n)-Θ (n+1) |＜ε), at this moment just can obtain " the special-purpose complexion model " of this specific image.

(4) aftertreatment (Post-Process): carry out last with " special-purpose complexion model " and take turns the skin area extraction, through denoising, generate comparatively complete and quite accurate skin area figure, at this moment still can exist some colour of skin distortions than higher skin shade, gauffer zone omission, cause among the area of skin color figure space to occur.On this figure, set up at last " border potential field ", give those spaces lower potential energy, flow to the lower place of potential energy in order to the guiding skin area and fill up these spaces, finally obtain complete, have other human body skin area of higher semantic level figure.

2. increment type gauss hybrid models (Incremental Gaussian Mixture Model) algorithm

The intention of increment type mixed Gauss model is based on following imagination: wish in online treatment image process system can dynamic modeling at the complexion model of specific image, the statistical information of " but general colour model " is not thought again to lose fully, promptly only wishes to utilize the sample set that newly obtains from specific image to revise on the basis of former complexion model.Yet can only obtain the parameter set of " general colour model " in the online treatment process, the original training sample collection is lost, and carries out the double sampling training so can't merge new samples collection and old sample set; So, have only a kind of new algorithm that can utilize the new samples collection on former mixed Gauss model parameter, to revise of design.[4,8,23] all can't satisfy prerequisite three conditions of this new algorithm simultaneously though can both be used for the study of gauss hybrid models:

(1) do not need double sampling.The increment type mixed Gauss model can not use the original sample set and the training again together of new samples collection of training " general colour model ", can only carry out online correction to the parameter of master pattern with newly advancing sample set with the pattern of increment.

(2) time complexity is low.Cannot use the algorithm of high time complexity, as the EM algorithm.

(3) factor of influence is set.Can artificially determine newly to add sample set to newly setting up the capability of influence of complexion model.

How description now designs the new algorithm that meets above three conditions.Suppose now the parameter set Θ=(w that possesses " general colour model " _m, μ _m, σ _m, 1≤m≤M) and new samples collection X={x _n, 1≤n≤N}, wherein M is the Gauss model number in the mixture model, N is that new samples is concentrated total sample number.In order to satisfy first condition, (Θ, X), wherein Θ ' is for through the new model after the increment type modeling need to be constructed as follows function the Θ '=F of form.Because increment type gauss hybrid models algorithm requires the increment type line modeling, this just requires this algorithm can not use batch mode (as the EM algorithm), and obviously, the most suitable online treatment of iterative scheme is decomposed into iterative formula to function F: Θ ⁽ⁿ⁺¹⁾=f (Θ ⁽ⁿ⁾, x _N+1), wherein subscript (n+1) and (n) the employed total sample number of expression this gauss hybrid models of training.The implication of this formula is: a given gauss hybrid models Θ who becomes with n sample training ⁽ⁿ⁾, as n+1 sample x _N+1To Θ ⁽ⁿ⁾Obtain new mixture model Θ after revising ⁽ⁿ⁺¹⁾Suppose that " general colour model " formed by K sample off-line training, there be N new samples to carry out online correction now to master mould, function F only needs N iteration just can build up " special-purpose complexion model " by increment, and time complexity is linear, has satisfied second condition.This algorithm can't double sampling special nature (original sample is lost) make the user manual intervention newly add the capability of influence of sample set on the contrary to new established model, very lucky, this is just meeting the 3rd condition.At first must clearly define factor of influence (Effect Factor) λ now, the number N that factor of influence can think newly to add sample set occupies the ratio of the total sample number (K+N) of training new model, i.e. N/ (K+N).Yet K and N all be objective definite, can't change, the factor of influence of expectation often can't obtain.In the training of " general colour model ", the sample set quantity of employing is very huge (1,500 ten thousand samples), i.e. K=15,000,000, and during online dynamic correction, initiate sample often wishes that the influence power to new model can reach half even more (λ 〉=0.5).Yet the online new samples collection order of magnitude that obtains often can only reach 10 ³～10 ⁴, N/ (K+N) is an extremely little numerical value, can't reach the influence power desired to new model far away, visible traditional double sampling algorithm does not possess the ability of regulating factor of influence.

Can change a kind of angle in " increment type gauss hybrid models " algorithm considers: " general colour model " Θ that obtains in advance ^(K)It is the one group of gauss hybrid models parameter set that forms with K sample training, though it is formed by K sample training, but it only is one group of master pattern, can think equally it by K ' (K '＜＜K) individual sample match forms, so can calculating K ', make it with N satisfy factor of influence λ: λ=N/ (K '+N), K '=N/ λ-N just can be expressed as Θ to " general colour model " then ^{(K ')}

Increment type gauss hybrid models algorithm can be expressed as " general colour model " Θ now ^(K), new samples collection X and factor of influence λ function: Θ ^(K+N)=F (Θ ^(K), X, λ), cancellation λ gets Θ ^{(K '+N)}=F (Θ ^{(K ')}, X), be the prototype of " increment type gauss hybrid models " algorithm.To utilize the above basic prototype derivation of equation to go out the concrete formula of this delta algorithm now.Below be the formulation of problem: suppose that given one K skin pixel sample off-line obtains with the training of EM algorithm, the Gauss model number is " the general colour model " of M (BIC determine), then the weight of m Gauss model, mean vector, covariance matrix respectively symbolically be w _m, μ _m, C _m, 1≤m≤M wherein.Having N newly to advance colour of skin sample now need revise " generally " complexion model, and the influence power expectation value is λ, and then the parameter set of new model can be expressed as:

$\left\{\begin{array}{c}{\mathrm{\Θ}}^{\left(K^{\′}\right)}=(w_m^{\left(K^{\′}\right)},{\stackrel{\→}{\mathrm{\μ}}}_m^{\left(K^{\′}\right)},C_m^{\left(K^{\′}\right)},1\≤m\≤M)\\ w_m^{(n+1)}=f_w(w_m^{\left(n\right)},x_{n+1}),K^{\′}\≤n{\≤K}^{\′}+N+1\\ {\stackrel{\→}{\mathrm{\μ}}}_m^{(n+1)}=f_{\mathrm{\μ}}\left({{\stackrel{\→}{\mathrm{\μ}}}_m^{\left(n\right)},x}_{n+1}\right),K^{\′}\≤n{\≤K}^{\′}+N-1\\ C_m^{(n+1)}=f_C(C_m^{\left(n\right)},x_{n+1}),K^{\′}\≤n{\≤K}^{\′}+N-1\end{array}\right.---\left(1\right)$

In order further to derive the iterative computation formula of " increment type gauss hybrid models " algorithm, the basic calculating formula of necessary clear and definite limited gauss hybrid models weight, mean vector, covariance matrix.Before this, at first need to define n sample x _nPosterior probability (degree of membership) for m Gauss model:

$P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})=\frac{w_m^{\left(Z\right)}{p}_m\left(x_n\right|{\mathrm{\Θ}}_m^{\left(Z\right)})}{\underset{j=1}{\overset{M}{\mathrm{\Σ}}}{w}_j^{\left(Z\right)}{p}_j\left(x_n\right|{\mathrm{\Θ}}_j^{\left(Z\right)})}---\left(2\right)$

1≤m≤M wherein, p _m() is the probability density function of m Gauss model, do note here: subscript Z is the used total sample number of this gauss hybrid models of training! Then:

$w_m^{\left(Z\right)}=\frac{\underset{n=1}{\overset{Z}{\mathrm{\Σ}}}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}{Z}---\left(3\right)$

${\stackrel{\→}{\mathrm{\μ}}}_m^{\left(Z\right)}=\frac{\underset{n=1}{\overset{Z}{\mathrm{\Σ}}}{x}_{n}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}{\underset{n=1}{\overset{Z}{\mathrm{\Σ}}}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}---\left(4\right)$

$C_m^{\left(Z\right)}=\frac{\underset{n=1}{\overset{Z}{\mathrm{\Σ}}}(x_n-{\stackrel{\→}{\mathrm{\μ}}}_m^{\left(Z\right)}{(x_n-{\stackrel{\→}{\mathrm{\μ}}}_m^{\left(Z\right)})}^{T}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)}))}{\underset{n=1}{\overset{Z}{\mathrm{\Σ}}}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}---\left(5\right)$

Below carry out the derivation of " increment type gauss hybrid models " algorithm iteration formula, all formula are all based on identical hypothesis Θ ^(Z+1)≈ Θ ^(Z)(when sample number Z enough big).The weight of m Gauss model, mean vector, covariance matrix are respectively in the gauss hybrid models:

$w_m^{(Z+1)}=\frac{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{(Z+1)})}{Z+1}\≈\frac{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}{Z+1}=\frac{Z{w}_m^{\left(Z\right)}+P\left(m\right|x_{Z+1},{\mathrm{\Θ}}_m^{\left(Z\right)})}{Z+1}---\left(6\right)$

${\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)}=\frac{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}{x}_{n}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{(Z+1)})}{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{(Z+1)})}\≈\frac{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}{x}_{n}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}=\frac{Z{w}_m^{\left(Z\right)}{\stackrel{\→}{\mathrm{\μ}}}_m^{\left(Z\right)}+x_{Z+1}P\left(m\right|x_{Z+1},{\mathrm{\Θ}}_m^{\left(Z\right)})}{Z{w}_m^{\left(Z\right)}+P\left(m\right|x_{Z+1},{\mathrm{\Θ}}_m^{\left(Z\right)})}---\left(7\right)$

$\begin{array}{c}{C}_m^{(Z+1)}=\frac{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}(x_n-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)}){(x_n-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)})}^{T}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{(Z+1)})}{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{(Z+1)})}\≈\frac{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}(x_n-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)}){(x_n-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)})}^{T}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}{\underset{n=1}{\overset{Z+1}{\mathrm{\Σ}}}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}\\ =\frac{\underset{n=1}{\overset{Z}{\mathrm{\Σ}}}(x_n-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)}){(x_n-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)})}^{T}P\left(m\right|x_n,{\mathrm{\Θ}}_m^{\left(Z\right)})}{Z{w}_m^{\left(Z\right)}+P\left(m\right|x_{Z+1},{\mathrm{\Θ}}_m^{\left(Z\right)})}+\frac{(x_{Z+1}-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)}){(x_{Z+1}-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)})}^{T}P\left(m\right|x_{Z+1},{\mathrm{\Θ}}_m^{\left(Z\right)})}{Z{w}_m^{\left(Z\right)}+P\left(m\right|x_{Z+1},{\mathrm{\Θ}}_m^{\left(Z\right)})}\\ =\frac{Z{w}_m^{\left(Z\right)}(C_m^{\left(Z\right)}+{\stackrel{\→}{\mathrm{\Γ}}}_m^{(Z+1)}{\stackrel{\→}{\mathrm{\Γ}}}_m^{{(Z+1)}^T})}{Z{w}_m^{\left(Z\right)}+P\left(m\right|x_{Z+1},{\mathrm{\Θ}}_m^{\left(Z\right)})}+\frac{(x_{Z+1}-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)}){(x_{Z+1}-{\stackrel{\→}{\mathrm{\μ}}}_m^{(Z+1)})}^{T}P\left(m\right|x_{Z+1},{\mathrm{\Θ}}_m^{\left(Z\right)})}{Z{w}_m^{\left(Z\right)}+P\left(m\right|x_{Z+1},{\mathrm{\Θ}}_m^{\left(Z\right)})}\end{array}----\left(8\right)$

At last, the iterative formula that is used for complete " the increment type gauss hybrid models " of the online increment modeling of Gauss model can be summarised as a circulation with false code:

for?Z＝K’：(K’+N-1)

Use above 3 formula (6), (7), (8) to calculate weight, mean vector, the covariance matrix of m Gauss model successively;

end

3. based on region growing (the Edge-based Region Growing) algorithm at edge

For certain specific image, it originates from " main seed region " based on the region growing at edge for the first time." main seed region " is to extract as follows: the window of one 16 * 16 pixel (is under 256 * 256 the situation in the image size, the size of window can be according to testing image size bi-directional scaling) on the initialization skin area figure that extracts by " general colour model ", move, if the skin area in the zone that this window covered (16 * 16 squares) accounts for 100%, then this zone is marked as " main seed region ", and all window areas that do not reach standard will be left in the basket.Extract through " main seed region ", just can be from these source regions based on the region growing at edge.

Seed region growth (Seeded Region Growing) is a proven technique, yet the region growing strategy in this algorithm also must the CONSIDERING EDGE constraint.Jointing edge information and region growing are used for also comparative maturity of Study of Image Segmentation, as [3,7,14], however these complicated dividing algorithms and be not suitable for cut apart (extraction) of skin area.In view of this, a kind of new algorithm that combines two kinds of classical technology arises at the historic moment, and it combines Canny operator [2] and Adam seed region growth [1], and the step of this algorithm is described below with false code:

(1) the unmarked pixel of adjacency of all pixels in " main seed region " is pushed sequential queue;

(2) while (sequential queue is not empty)

(3) from formation, take out first pixel v;

(4) with v be the center covers one 5 * 5 pixel on image window W;

(5) among the calculation window W the underlined YUV average vector x that crosses pixel;

(6) the YUV vector of calculating pixel v;

(7) if (| x-y|＜δ and do not have the edge to pass window W)

(8) v is labeled as skin pixels;

(9) adjacency of v unmarked (not belonging to " main seed region ") pixel is pushed sequential queue;

(10)end

Wherein || the expression Euclidean distance, δ gets empirical value, generally chooses in the 20-30 scope.

4. border potential field (Boundary Potential Field) algorithm

" general colour model " being revised after convergence, just can obtain " special-purpose complexion model " at certain specific image S through too much taking turns " increment type gauss hybrid models " algorithm.With this model image S is carried out last area of skin color and extracts, much accurate, complete skin area figure in the time of just can obtaining an amplitude ratio and use " general colour model ", yet several point defects below should figure still existing:

(1) human body skin gauffer can occur unavoidably, the each several part of human body self structure also can cover mutually and cause shade, these all can cause the generation in color change zone on the skin, the area of skin color growth is owing to rely on color distortion (Euclidean distance) to retrain, these color changes can cause the area of skin color can't normal growth, occur random hollow out, slit, disconnection in area of skin color figure.

(2), thereby sketched the contours beautiful area of skin color border though the canny edge has effectively stopped the area of skin color further growth; But because the limitation of canny operator, promptly the edge can't form closed loop and have a large amount of irregular funiclar curves, causes the skin area irregularity boundary at obscure place, edge in area growth process.

(3) in specific image often human body skin area and background owing to color too near causing the canny operator can't detect the edge of the colour of skin and background intersection, can't rely on the edge to tackle region growing, can only rely on color to retrain purely.At this moment " based on the region growing at edge " deteriorates to traditional " seed region growth (Seeded RegionGrowing) ".

Can make following guess based on a large amount of observationss: in area of skin color figure, very likely remain skin area outside the border of those random hollow out zones and distortion significantly, but because color distortion is too big in its color and the area of skin color, be subjected to the constraint of color distance, cause this zone can't continue to outgrowth; The smooth area of skin color in those borders then very likely is outward a background, but because background color may be more approaching with color in the area of skin color, will grow to the background direction in the zone.Draw the thought of foundation " border potential field " now: outside the border of random hollow out zone and distortion significantly, give lower potential energy, outside large tracts of land smooth region border, give higher potential energy, giving absolute high potential energy away from the area of skin color place, allow the color degree of approach and the acting in conjunction of potential energy intensity, area of skin color then can continued growth when both acting in conjunction power reaches a certain standard.This imagination can be formulated as:

D (the color degree of approach) * E (potential energy intensity)≤C (constant) (9)

The physical significance of this formula is:

(1) color of new growth part is approaching with the area of skin color color, area of skin color border external potential energy a little less than, then might remain area of skin color more outside the area of skin color border, can without the least hesitation grow in this zone.

(2) color and the area of skin color color of new growth part are approaching, but area of skin color border external potential energy is very strong, are background probably outside the area of skin color border then, as long as both actings in conjunction can also be grown less than this zone of constant.

(3) color and the area of skin color color distortion of new growth part are bigger, but area of skin color border external potential energy is very weak, are skin probably outside the area of skin color border then, as long as both actings in conjunction can also be grown less than this zone of constant.

(4) color and the area of skin color color distortion of new growth part are big, and area of skin color border external potential energy is strong, then might be background more outside the area of skin color border, and both actings in conjunction are easy to surpass constant, and this zone is difficult to continued growth.

Define the border potential field now.Given set U={v _i(x _i, y _i), wherein i is the pixel index among the skin area figure, and v _iNot in the skin area that is partitioned into., for each element v among the set U _i, be the center window of placing a h * h (W ') with it, the limited area among the window W ' just can be thought a plastid pockety like this, and the pixel of the middle skin area of window W ' can be thought the particle m of unit.Be the pixel in any non-area of skin color among the window W ', promptly belong to the element v among the set U _iGive potential energy, potential energy assignment principle is all based on this supposition: with v _iIn the fixed size window for the center, if the big and area of skin color pixel coordinate of the shared area ratio of area of skin color approaches v in the window _i, then this potential energy is low.For this reason, the length of side that window W ' at first is set is h=27, with v _iSet up two-dimentional rectangular coordinate system for initial point, be subjected to the inspiration of the law of universal gravitation, v _iThe potential energy at place can be defined as:

For the calculability of potential energy intensity and with the comparability of color similarity degree, must the cancellation following formula in gravitational constant G, regular potential energy intensity E simultaneously:

$E\left(W^{\&prime;}\right)=\underset{x=-h/2}{\overset{h/2}{\mathrm{\&Sigma;}}}\underset{y=-h/2}{\overset{h/2}{\mathrm{\&Sigma;}}}\frac{\mathrm{Gm}\&CenterDot;1}{\sqrt{{\mathrm{<figure-callout\; id="2"\; label="x\; i"\; filenames="A20051002707800171.png"\; state="\{\{state\}\}">x</figure-callout>}}_i^{}+y_i^2}},(x_i,y_i)\&Element;W^{\&prime;}---\left(11\right)$

E′(v _i)＝255·exp{-E(v _i)/E(W′)} (12)

Can directly reflect (promptly having comparability with the color similarity degree) with color through the above potential energy effect of handling among the skin area figure of back: the whole white zone is absolute high potential energy district in the drawings, and how new growth part color and area of skin color color are near all continued growths again; The grey black look zone on skin area border is expressed as the low-potential energy district, and color is black represents potential energy low.

In the self-adaptation framework, algorithm much at one for drainage in the post-processing module (region growing) strategy and " based on the region growing at edge ", unique difference just is in " based on the region growing at edge " algorithm the judgment expression in the 7th step---the edge no longer considers, and change in potential energy.Follow 5 * 5 window W, for pixel v all among the SSL, its color similarity degree E _vWith potential energy D _vCan be defined as:

$D_v=\sqrt{\frac{{(y_v-\stackrel{\_}{y})}^2+{(u_v-\stackrel{\_}{u})}^2+{(v_v-\stackrel{\_}{v})}^2}{3}}---\left(14\right)$

In the above in second formula, two three-dimensional YUV vectors are respectively the average color proper vector of all skin pixels among the color feature vector of pixel v and the window W.At last, the judgment expression in the 7th step can be revised as If (G _v* E _v≤ δ), wherein δ gets empirical value, generally can in the 600-1200 scope, choose.

Application example

Accompanying drawing 2 is demonstrated the concrete steps of above-mentioned algorithm flow framework by a concrete example, by provide each algoritic module in the middle of the final segmentation result of output and this framework, give intuitively and understand.Wherein (a) is the color digital image that has human body skin area of input; (b) skin area that extracts for " general colour model ", wherein white line is the Canny edge, the shadow region in the skin area is " a main seed region "; (c) be " based on the region growing at edge " effect, obtain the shadow region among the figure to outgrowth from " main seed region " beginning; (d) skin area for extracting with " special-purpose complexion model "; (e) be foundation " edge potential field " on the skin area figure that " special-purpose complexion model " extracts, gray level expressing potential energy intensity, black more potential energy is low more; (f) the human body skin area figure for finally being partitioned into.

Claims (6)

1. A method for segmenting human skin regions in color digital images and videos, characterized in that the specific steps are as follows: (1) Offline processing Manually extract a large number of general skin color pixel samples from a large number of images containing different human skin regions, and use the EM algorithm to establish a "general skin color model"; (2) Preprocessing Detect the Canny edge from the image to be tested, use the "general skin color model" to extract the "main seed area" as the growth source, perform edge-based area growth, and finally perform light source compensation; (3) Incremental processing Perform "edge-based region growth" from the main seed area to obtain specific skin color pixel samples in the image, apply the incremental Gaussian mixture model algorithm to correct the online skin color model, and use the newly obtained model from Extract the skin area from the image, perform area growth, correct it online, and cycle until the skin color model converges to obtain a "special skin color model" for the specific image; (4) Post-processing Use the "special skin color model" to extract the skin area, generate an accurate skin area map, and finally establish a "boundary potential field" on the map to guide the skin area to flow to places with lower potential energy to fill in due to shadow folds, etc. Gaps in the resulting skin area. 2. The method according to claim 1, wherein the prototype of the incremental Gaussian mixture model algorithm is: Θ ^(K'+N) =F(Θ ^(K') , X), where Θ is The Gaussian mixture model parameter set, the superscript (K'+N) and (K') indicate that the Gaussian mixture model Θ is trained by K'+N and K' samples, X is the new sample set, and the function F indicates the The original skin color model Θ ^(K') and the new sample set X are incrementally modeled to obtain a new skin color model Θ ^(K'+N) . 3. The method according to claim 2, characterized in that the iterative calculation formula of the incremental Gaussian mixture model algorithm is: ${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; Zw</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&Theta;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}$ ${\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; Z</span>}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&Theta;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; Zw</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&Theta;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}$ ${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; Z</span>}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&Gamma;</span>}}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&Gamma;</span>}}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; Zw</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&Theta;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>\frac{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&Theta;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; Zw</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&Theta;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}$ Among them, w _m ^(Z+1) , μ _m ^(Z+1) , and C _m ^(Z+1) are the weight and mean vector of the mth Gaussian model in the Gaussian mixture model trained with Z+1 samples, respectively , the iteration value of the covariance matrix, where $\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&Theta;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; \&Theta;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}{\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; \&Theta;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}$ is the posterior probability of the nth sample relative to the mth Gaussian model, where 1≤m≤M, p _m (·) is the probability density function of the mth Gaussian model. 4. The method according to claim 1, wherein the main seed area is extracted as follows: a window of 16×16 pixels moves on the initial skin area map extracted by the “general skin color model”, if If the skin area in the area covered by this window accounts for 100%, then this area is marked as the "main seed area", and all window areas that do not meet the standard will be ignored. 5. The method according to claim 1, characterized in that the steps of the edge-based region growing algorithm are described in pseudocode as follows: (1) Push the adjacent unmarked pixel points of all pixels in the "main seed area" into the sequence queue; (2) while (sequential queue is not empty) (3) Take out the first pixel v from the queue; (4) Overlay a window W of 5×5 pixels on the image with v as the center; (5) Calculate the YUV average vector x of all marked pixels in the window W; (6) Calculate the YUV vector of pixel v; (7) if(|x-y|<δ and no edge passes through the window W) (8) Mark v as a skin pixel; (9) Push the adjacent unmarked (not belonging to the "main seed area") pixel point of v into the sequence queue; (10) end Where |·| represents the Euclidean distance, and δ is an empirical value. 6. The method according to claim 1, characterized in that the algorithm for establishing the boundary potential field is as follows: a given set U={v _i (xi _, y _i ), wherein i is a pixel in the skin area map Index, and v _i is not in the segmented skin area. }, for each element v _i in the set U, place a h×h window W' centered on it, so that the limited area in the window W' is regarded as a plastid with uneven distribution, and the skin in the window W' The pixel in the area can be considered as the unit mass point m; for the pixel in any non-skin color area in the window W', that is, the element v _i belonging to the set U is endowed with potential energy. For this reason, the side length of the window W' is first set to be h= 27. Establish a two-dimensional Cartesian coordinate system with v _i as the origin, and the potential energy at v _i is defined as: Eliminate the gravitational constant G in the above formula, and normalize the potential energy intensity E at the same time: $\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; W</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 2</span>}}{\overset{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 2</span>}}{\overset{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\frac{\mathrm{<span\; class="notranslate">\; G\; m</span>}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\sqrt{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; W</span>}}^{<span\; class="notranslate">\; \&prime;</span>}$ E'(v _i )=255·exp{-E(v _i )/E(W')}.

Images