CN1508756A - Sensitive image identifying method based on body local and shape information - Google Patents

Abstract

The invention is a kind of sensitive image incrimination method based on local and body information. The steps are: divides the static sensing image; determines the region geometric point; uses the autonomous bounce of growing point to determine the image local character of growing point region; carries on image property judgment.

Description

Sensitive image recognition methods based on body local and body information

Technical field

The present invention relates to area of pattern recognition, particularly based on the sensitive image recognition methods of body local and body information.

Background technology

Along with the develop rapidly of modern internet technology, network is to the infiltration of global economy and social life, and its degree of depth and influence power be people's expectation head and shoulders above.The network information security becomes a very important problem gradually, wherein to society, especially pupillary influence is more caused people's extensive concern, so the information filtering technology has become urgent theory and actual demand.In the U.S., these problems have just caused the public's attention as far back as 1994, American society was subjected to deeply easily that the misery of harmful network informations such as the online porny of network, violence, vicious speech perplexs at that time, and many news, newspaper, magazine all are flooded with the fear to problems such as porn site, various ugly group, online assaults sexually.A large amount of harmful contents directly causes Congress to pass through " CommunicationsDecency Act (CDA) " and " Child On-line Protection Act (COPA) " two laws on the network.As legal basis, the software industry of the U.S. has been developed themselves Web content filter software, and sets up Web content auditing system platform.Passed through " Childrens ' Internet Protection Act (CIPA) " law subsequently again in Congress in 1999 to protect young people, made it avoid the infringement of network harmful information.

To the sensitive information context of detection, abroad some universities (Berkeley, Iowa Standford) have carried out the exploration that part is analyzed sensitization picture on the network.Fleck and Forsyth be by the skin of human body, and the each several part skin area is linked to be one group, discerns a width of cloth picture and whether comprise bare content.James Ze Wang utilizes WIPE (Wavelet Image PornographyElimimation) method that sensitization picture is discerned and filtered.This method synthesis has utilized the Daubechies wavelet transformation, normalization central moment, and color histogram forms the semantic matches vector and carries out image classification identification.The Ian Craw of Aberdeen learns the probability model of the colour of skin with the SOM net in skin detection, obtains one behind the test samples fan-in network and may be the probable value of the colour of skin, a threshold values is set then takes a decision as to whether the colour of skin.In addition, also have some general CBIR systems, as the QBIC of IBM, the ImageFinder of Attrasoft, the Imatch of MWLabs etc.What deserves to be mentioned is that four scientists of French Inst Nat Rech Inf Automat image and multimedia index group have set up LookThatUp company in 1999, the said firm's image filtering in industry maintains the leading position with the retrieval product.LookThatUp's.Image-FilterTM utilizes advanced recognizer to carry out real time filtering to the image on the network.

In calendar year 2001, Europe starts the NetProtect plan, and this plan is since end of day in 1 day to 2002 May 1 of January calendar year 2001, by the EADS Matra Systemes of France; Information research institution unites the Matra Global Netservices of Hispanic Red Educativa, France, the Hyertech of Greece, the scientific research institutions such as SailLabs of Germany develop jointly.The target of NetProtect plan is to set up the uniform technical standards of european internet information filtering instrument, to realize cross-region, to stride the internet harmful information filtration of language.

Domestic existing anti-yellow software has U.S. duckweed software work chamber to release the anti-yellow bodyguard in the Forbidden City of the anti-yellow expert of U.S. duckweed, ZiJinCheng.NET release, flies the Escort who releases great waves software work chamber, the anti-yellow software of " piercing eye " computer of news Fetion breath Science and Technology Ltd. of China Science ﹠ Technology University exploitation, anti-yellow software of " five-element bodyguard " computer of Tsing-Hua University or the like.What need proposition is no matter these domestic network harmful information filtration softwares all can not reach due effect technically or from filter method.The particularly develop rapidly of China's network application in the last few years causes network far-reaching day by day to the influence of society, family, education, so the network harmful information filtration will face unprecedented pressure.

It is emphasized that, though internet harmful information filtration technology has worldwide obtained paying close attention to widely and studying, but still have many difficult points aspect the harmful information recognition technology, wherein porny identification and the filter method based on picture material still lacks effective algorithm and sorting technique.Therefore how to develop more robust, the sensitive image recognition technology is still a challenge accurately.

Summary of the invention

The purpose of this invention is to provide a kind of part that utilizes the human body that sensitization picture can express and body information and reach method the identification of sensitive image.

For achieving the above object, the sensitive image recognition methods based on body local and body information comprises step:

Static sensitive image is divided;

Determine the region geometry point;

The image local feature of growing point affiliated area is determined in the autonomous shake of employing growing point;

Carrying out image property judges.

The present invention is a kind of novel sensitive image recognition technology, broken through such as an international difficult problem that has aspects such as sensitive image recognition technology speed is slow, efficient is low, device dependence is strong now such as color histogram coupling, wavelet transformation outline, skin tone texture description, centralized moments of image coupling, had broad application prospects.

Description of drawings

Fig. 1 is a zoning plan;

Fig. 2 is the area attribute graph of a relation;

Fig. 3 growth district and candidate region graph of a relation;

Fig. 4 is a direction of growth and growth pattern figure;

Fig. 5 judges flow process.

Embodiment

Principal feature of the present invention is: 1) taked a kind of novel image general characteristic decision method, promptly regional preliminary judgement method.It can tentatively judge the character of image according to the feature of the relevant range of choosing in the image, and this will help the processing of subsequent process; 2) establish corresponding geometric point on the region base of determining, the growth pattern based on the colour of skin has been taked in the shake growth of geometric point, and this mode has shortened the time of geometric point growth greatly, has reduced calculation cost; 3) on the basis of colour of skin point location, set up the geometric relationship between the colour of skin point.Under this geometric relationship, establish image information acquisition method between points.This process has not only been obtained the local message of image, also gives expression to corresponding body information simultaneously; 4) on the basis of the image information of obtaining, establish the image fast classification method.

Provide the explanation of each related in this invention technical scheme detailed problem below in detail.

The image geometry area dividing

The image geometry area dividing is to determine the starting condition of characteristics of image, and when dividing, algorithm has also extracted the geometric relationship between the zone.If image is that (x, y), each zone is f to f _i(x _i, y _i) then the relation between them can be expressed as follows

f(x，y)＝f ₁(x ₁，y ₁)+f ₂(x ₂，y ₂)+f ₃(x ₃，y ₃)+…+f _n(x _n，y _n) (1)

N=16 in the formula.Image is through dividing subregion f _i(x _i, y _i) represented local message partly to a certain extent, division methods is seen accompanying drawing 1.On the basis of this division, set up the algebraic relation between image region and the growth district, matrix representation is as follows:

A is that the image-region attribute is determined matrix, and P is the growing point attribute matrix, a ' _IjBe that image region belongs to

$A=\left[\begin{array}{cccc}{a}_{11}& a_{12}& a_{13}& a_{14}\\ a_{21}& a_{22}& a_{23}& a_{24}\\ a_{31}& a_{32}& a_{33}& a_{34}\\ a_{41}& a_{42}& a_{43}& a_{44}\end{array}\right]$ $P=\left[\begin{array}{ccc}{p}_{11}& p_{12}& p_{13}\\ p_{21}& p_{22}& p_{23}\\ p_{31}& p_{32}& p_{33}\end{array}\right]$

$a_{\mathrm{ij}}^{\′}=\left[\begin{array}{cc}{p}_{(i-1)(j-1)}^{\left(4\right)}& p_{(i-1)j}^{\left(3\right)}\\ p_{i(j-1)}^{\left(2\right)}& p_{\mathrm{ij}}^{\left(1\right)}\end{array}\right]$ $p_{\mathrm{ij}}^{\′}=\left[\begin{array}{cc}{p}_{\mathrm{ij}}^{\left(1\right)}& P_{\mathrm{ij}}^{\left(2\right)}\\ p_{\mathrm{ij}}^{\left(3\right)}& p_{\mathrm{ij}}^{\left(4\right)}\end{array}\right]$

The property judgment matrix, p ' _IjBe growing point determined property matrix, wherein p _Ij ^(e)It is growth district under the growing point.So just set up the mutual relationship between image region, growing point, the growth district, expression directly perceived can be with reference to figure 2, Fig. 3.

Region geometry point is established and the shake growth

Get the point of crossing p in four adjacent subarea territories _IjAs how much representative points in zone, the local feature that how much shake growth will be expressed the information of growing point affiliated area.For sensitive image, the prominent feature that image had is exactly the expressed content of the colour of skin, if (r, g b) represent colour of skin information, function G with function F _Ij(F (r, g, b), x _Ij, y _Ij) be geometric point p _IjGrowth function.Then for any one zone wherein, the mathematical description of geometric point growth pattern is as follows

$p_{\mathrm{ij}}\∈p_{\mathrm{ij}}^{\left(e\right)}{\⇔G}_{\mathrm{ij}}(F(r,g,b),x_{\mathrm{ij}}^{\left(e\right)},y_{\mathrm{ij}}^{\left(e\right)})-----\left(2\right)$

The constraint condition that growth stops

F (r, g, b)≤and α, α is given threshold value.(3)

Expression directly perceived is seen accompanying drawing 4.The numerical value that shake growth and determined property by region geometry point can obtain each element among matrix A, the P.Work as α _Ij=0 o'clock, respective regions was non-candidate region in the correspondence image, works as α _Ij=1 o'clock, respective regions was the candidate region in the correspondence image, and the formation of human body body just relies on the determined candidate region of entire image in the image.Work as p _Ij=0 o'clock, this growing point was non-candidate point, works as p _Ij=1 o'clock, this growing point was a candidate point, and the formation of image body local feature just relies on the determined candidate point of growing point matrix.

The generation of local feature and body characteristics

By dividing region and growing point shake growth, we have obtained image candidate subregion and candidate's anchor point.In these zones, we extract contains the essential information of the more rich zone of colour of skin information as further computing.That is to say that the prime area is 16, judge that through the feature of previous step the size of the number of regions N that chooses now should be 0≤N≤9, N is an integer.Why number of regions N is because zone and point be closely related less than 9.Mathematical form is described below

I ' | i≤i '≤j, i ' ∈ integer } (4)

K=y wherein _j-y _i/ x _j-x _i, b=kx _i+ y _i

So S (kx _i'+b, x _{I '}) be straight line

On point set.To each some utilization growth function G among the point set S (F (and r, g, b), S _{I '}) just can obtain corresponding local feature.Should have for entire image $M=C_N^2$ Individual point set, wherein N 〉=2 should be so the local message of image is expressed function

$\mathrm{LocalFun}(x,y)=\mathrm{\ΣLocalFun}(f_i,G_{s_m})----\left(5\right)$

Image body information is the relation of the position between the point set in fact.When determining image local feature, algorithm images acquired point is concentrated the positional information of each point, through the feature extraction of each dot position information being formed the body information of human body content in the image.So body information representation function should be

$\mathrm{ShapeFun}(x,y)=\underset{i=0}{\overset{N}{\mathrm{\Σ}}}F\left(S_i(x_i,y_i)\right)-----\left(6\right)$

Wherein be linear F (S _i(x _i, y _i)) piecewise function.

Image property is expressed

Above-mentioned analysis provides the local message and the body information of image, and the combination of the two can obtain the description to image property.If it is FeatureFunO that characteristics of image is expressed function, then

FeatureFun(f(x，y)＝LocalFun(x，y)+ShapeFun(x，y) (7)

So for arbitrary width of cloth image, the local feature number of image should be

The body characteristics number should be N _S=1.And the image body number that this algorithm can be expressed should be

$N_S=C_9^0+C_9^1+C_9^2+C_9^3+C_9^4+C_9^5+C_9^6+C_9^7+C_9^8+C_9^9------\left(9\right)$

The image total characteristic number that algorithm can be expressed should be

N _F＝N _L*N _S (10)

Image property is judged

The expression to picture material has been finished in above-mentioned analysis, and this part will be judged image property according to expressed picture material; If with ClassFunO presentation video character discriminant function, then

The image property discriminant function is actually the image classification function, and when functional value was 1, image was a sensitive image, and when functional value was O, image was non-sensitive image.

Enantiomorph information adopts the geometric similarity degree to measure, and intuitively can be regarded as the similarity relation of plane triangle, and even two plane triangles are similar, and then their pairing angles equate.Formulation is as follows

But for the similar judgement of image body, algorithm does not require that body information is similar fully, satisfies given thresholding and gets final product.

For the distance metric method under the image local information employing body similarity relation, formulation is as follows

$\min D=\mathrm{\Σ}\sqrt{{(x-x^{\′})}^2+{(y-y^{\′})}^2}---\left(13\right)$

It is a lot of to satisfy perhaps having of body similarity relation, but satisfy local feature under the body similarity relation nearest have only one.Whole deterministic process is seen accompanying drawing 5.

In order to implement concretism of the present invention, the internet sensitive image filtering system that we design and have realized discerning based on picture material, this algorithm has embodied very high arithmetic speed, can satisfy the requirement of real time filtering fully, simultaneously we have designed the active recognition system based on image library retrieval identification, and the recognition correct rate of 13643 width of cloth images is reached more than 85%.

The whole recognizer of image is carried out routine call with the form of dynamic link library file.It resembles a water swivel that has screen pack, can filter out the objectionable impurities in the water, and different is that water swivel needs regularly to clean, and our picture filter does not need manual intervention, and it can abandon harmful network packet automatically.

Claims (4)

1. sensitive image recognition methods based on body local and body information comprises step:

Static sensitive image is divided;

The mutual relationship of candidate region and growth district;

Determine the region geometry point;

The image local feature of growing point affiliated area is determined in the autonomous shake of employing growing point;

Carrying out image property judges.

2. by the described method of claim 1, it is characterized in that described definite region geometry point comprises step:

Get how much representative points of the point of crossing in 4 adjacent subarea territories as the zone;

The geometric point growth pattern satisfies following formula:

$p_{\mathrm{ij}}\∈p_{\mathrm{ij}}^{\left(e\right)}{\⇔G}_{\mathrm{ij}}(F(r,g,b),x_{\mathrm{ij}}^{\left(e\right)},y_{\mathrm{ij}}^{\left(e\right)})$

3. by the described method of claim 1, it is characterized in that described zone is equal to or less than 9.

4. by the described method of claim 1, it is characterized in that described image property judgement comprises step:

In a plurality of image bodies that satisfy the body similarity relation, select nearest one of local feature.

Images