CN102163286B - Pornographic image evaluating method - Google Patents

Abstract

The invention discloses a pornographic image evaluating method. The method comprises the following steps of: dividing the complexion of an image; extracting 31-dimensional characteristics, namely the integral characteristics of the image, the local part characteristics of a human body and human body shape characteristics, to form characteristic vectors; and training an evaluation model, evaluating the pornographic degree of an input image, and outputting an evaluation result. By the method, whether the input image is a pornographic image or not can be judged, the pornographic degree of the image can be evaluated, 0 to 100 percent of evaluation results can be output, and a larger numerical value means the higher pornographic degree of the image, so that pornographic images is convenient to filter in a classified mode.

Description

A kind of Pornographic image evaluating method

Technical field

The present invention relates to the Computer Applied Technology field, particularly a kind of Pornographic image evaluating method.

Background technology

Current, in the whole nation even world wide, extensively carry out the special campaigns of the obscene pornographic of joint operation internet and mobile media and vulgar information.Because teen-age self-control is relatively poor, in the middle of juvenile deliquency, the temptation that has nearly 80% people to be the network flame.In order to help the teenager netizen to exempt from the erosion of pornographic information, reduce the probability that they go on the crime road, pornographic degree to image is carried out effective evaluation and filtration, technically these harmful informations is tackled before the teenager obtains them, just seems particularly necessary.

Mostly adopt the simple feature such as colour of skin ratio in the prior art scheme, fail to consider integral image feature, body local genius loci and trunk shape facility, thereby discrimination is lower, and mostly Output rusults is pornographic or normal logical value, can not measure the pornographic degree of image.

Summary of the invention

The technical matters that (one) will solve

In view of this, fundamental purpose of the present invention is that to solve the prior art discrimination lower, the problem that can not measure the pornographic degree of image, for this reason, the invention provides a kind of Pornographic image evaluating method, can estimate the pornographic degree of input picture, and the output evaluation result.

(2) technical scheme

For achieving the above object, the invention provides a kind of Pornographic image evaluating method, the method comprises:

Step 1: the colour of skin to input picture is cut apart, and obtains area of skin color;

Step 2: totally 31 dimensional feature constitutive characteristics are vectorial by global feature, body local genius loci and trunk shape facility three classes at the good great amount of images sample extraction input picture of handmarking;

Step 3: the evaluation of training model is also estimated the pornographic degree of input picture, output pornographic image evaluation result.

(3) beneficial effect

Can find out that from technique scheme the present invention has the following advantages:

Utilize the present invention, can judge not only whether input picture is pornographic image, and can estimate the pornographic degree of image that the present invention has improved discrimination, can measure the pornographic degree of image.

1, this Pornographic image evaluating method provided by the invention owing to considered integral image feature, body local genius loci and trunk shape facility, can carry out effectively evaluating to the pornographic degree of image effectively.

2, this Pornographic image evaluating method provided by the invention, because output is the evaluation result of the pornographic degree of image, and be not only simple judgement graphic sex whether, and export 0%～100% evaluation result, numerical value is larger to show that the pornographic degree of image is higher, is convenient to pornographic image is carried out classified filtering.

Description of drawings

Fig. 1 is the process flow diagram of Pornographic image evaluating method.

Fig. 2 is that the human body part detects and the trunk testing result

Embodiment

For making the purpose, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, the present invention is described in more detail.

Execution environment of the present invention is made of three module groups below the computer realization:

One, colour of skin extraction module, the major function of this module are that input picture is carried out skin color segmentation, obtain area of skin color.

Two, characteristic extracting module, the major function of this module are to extract integral image feature, body local genius loci and trunk shape facility three classes totally 31 dimensional feature constitutive characteristics vector at input picture.

Three, model training and evaluation module, the major function of this module are by extracting feature at the good great amount of images sample of handmarking, with these feature input evaluation models, to the evaluation model training, obtaining model parameter.The evaluation model that utilization trains is estimated input picture, and the output evaluation result.

Below in conjunction with Fig. 1, provide in detail the explanation of each related in technical solution of the present invention step details problem.

Step 101: to the input picture skin color segmentation.Can adopt arbitrarily skin color segmentation algorithm, mainly contain two class skin color segmentation algorithms, a class is pixel scale, another kind of be region class other.Owing to can consider the attributes such as texture of the colour of skin, thereby usually can obtain preferably segmentation effect based on the skin color segmentation algorithm in zone.

Step 102: extract integral image feature, body local genius loci and trunk shape facility three classes totally 31 dimensional feature constitutive characteristics vector.

One, integral image feature

The global feature of image is comprised of following 5 parts:

The image length breadth ratio;

The entropy of image;

The entropy of gray level image is defined as:

$\mathrm{Entropy}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{p}_i\·{\log }_2{p}_i---\left(1\right)$

In the formula, p _iThat gray-scale value is the number percent that the number of pixels of i accounts for the entire image sum of all pixels.

The colour of skin ratio of image;

Colour of skin ratio:

Colour of skin connected region number;

The geometric moment of area of skin color (this feature has 10 dimensions);

(p+q) rank geometric moment of image I (x, y) is defined as:

$M_{\mathrm{pq}}=\underset{x}{\mathrm{\Σ}}\underset{y}{\mathrm{\Σ}}{x}^p{y}^{q}I(x,y)---\left(3\right)$

In the formula, x presentation video lateral coordinates, y presentation video along slope coordinate, p, q is nonnegative integer arbitrarily.

We extract the front 3 rank geometric moments of area of skin color:

0 rank (p+q=0) geometric moment: M ₀₀

1 rank (p+q=1) geometric moment: M ₁₀, M ₀₁

2 rank (p+q=2) geometric moment: M ₂₀, M ₁₁, M ₀₂

3 rank (p+q=3) geometric moment: M ₃₀, M ₂₁, M ₁₂, M ₀₃

These 10 geometric moments are as the geometric moment feature of area of skin color, and the geometric moment character representation of this area of skin color is as follows: M ₀₀, M ₁₀, M ₀₁, M ₂₀, M ₁₁, M ₀₂, M ₃₀, M ₂₁, M ₁₂, M ₀₃

Two, body local genius loci

Next 4 features are relevant body local location detection (seeing Fig. 2), and we adopt the cascade classifier based on Ha Er (Haar) feature to carry out the body local location detection, and described 4 features are expressed as follows:

The people's face quantity that detects;

The chest quantity that detects;

The private section quantity that detects;

The ratio of the shared integral image colour of skin of face complexion;

Three, trunk shape facility

Last 7 features from maximum colour of skin UNICOM zone and fitted ellipse (seeing Fig. 2) extract, the trunk shape of human body has mainly been described, described 7 features are expressed as follows:

Maximum colour of skin UNICOM region area accounts for the ratio of total colour of skin area;

Maximum colour of skin UNICOM regional center is to the distance of picture centre;

The geometric invariant moment (this feature has 7 dimensions) in maximum colour of skin UNICOM zone;

The geometric invariant moment of image I (x, y) has the character such as rotation, translation, yardstick be constant.For the geometric invariant moment of computed image, at first need to introduce the center square μ of image _Pq:

${\mathrm{\μ}}_{\mathrm{pq}}=\underset{x}{\mathrm{\Σ}}\underset{y}{\mathrm{\Σ}}{(x-\stackrel{\‾}{x})}^p{(y-\stackrel{\‾}{y})}^{q}I(x,y)---\left(4\right)$

P wherein, q is nonnegative integer arbitrarily,

The barycenter of image-region, M ₁₀, M ₀₁Image moment for formula (3) definition.

Then center, the 0-3 rank square of image can be expressed as:

μ ₀₀＝M ₀₀，

μ ₀₁＝0，

μ ₁₀＝0，

μ ₁₁＝M ₁₁-xM ₀₁＝M ₁₁-yM ₁₀，

μ ₂₀＝M ₂₀-xM ₁₀，

μ ₀₂＝M ₀₂-yM ₀₁，

μ ₂₁＝M ₂₁-2xM ₁₁-yM ₂₀+2x ²M ₀₁， (5)

μ ₁₂＝M ₁₂-2yM ₁₁-xM ₀₂+2y ²M ₁₀，

μ ₃₀＝M ₃₀-3xM ₂₀+2x2M ₁₀，

μ ₀₃＝M ₀₃-3yM ₀₂+2y2M ₀₁.

The center square of image only has translation invariant character.For the situation of p+q 〉=2, can adopt following formula construction to have the square η of yardstick invariance _Pq:

${\mathrm{\η}}_{\mathrm{pq}}=\frac{{\mathrm{\μ}}_{\mathrm{pq}}}{{\mathrm{\μ}}_{00}^{(p+q)/2+1}}---\left(6\right)$

On this basis, we extract following geometric invariant moment as feature, this feature h in the maximum colour of skin of image UNICOM zone ₁To feature h ₇As shown in the formula illustrating:

h ₁＝η ₂₀+η ₀₂，

h ₂＝(η ₂₀-η ₀₂) ²+(2η ₁₁) ²，

h ₃＝(η ₃₀-3η ₁₂) ²+(3η ₂₁-η ₀₃) ²，

h ₄＝(η ₃₀+η ₁₂) ²+(η ₂₁+η ₀₃) ²，

h ₅＝(η ₃₀-3η ₁₂)(η ₃₀+η ₁₂)[(η ₃₀+η ₁₂) ²-3(η ₂₁+η ₀₃) ²]+

(3η ₂₁-η ₀₂)[3(η ₃₀+η ₁₂) ²-(η ₂₁+η ₀₃) ²]，

h ₆＝(η ₂₀-3η ₁₂)[(η ₃₀+η ₁₂) ²-3(η ₂₁+η ₀₃) ²]+4η ₁₁(η ₃₀+η ₁₂)(η ₂₁+η ₀₃)

h ₇＝(3η ₂₁-3η ₀₃)(η ₃₀+η ₁₂)[(η ₃₀+η ₁₂) ²-3(η ₂₁+η ₀₃) ²]+ (7)

(η ₃₀-3η ₁₂)(η ₂₁+η ₀₃)[3(η ₃₀+η ₁₂) ²-(η ₂₁+η ₀₃) ²].

The circularity in maximum colour of skin UNICOM zone;

We adopt following formula to calculate the circularity R in maximum colour of skin UNICOM zone:

$R=\frac{{\left|\right|F_1\left|\right|}^2}{{\left|\right|F_{-1}\left|\right|}^2}---\left(8\right)$

Wherein, $F_1=\frac{1}{N}(\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}(x_k\cos \left(\frac{2\mathrm{\πk}}{N}\right)+y_k\sin \left(\frac{2\mathrm{\πk}}{N}\right))+j\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}(y_k\cos \left(\frac{2\mathrm{\πk}}{N}\right)-x_k\sin \left(\frac{2\mathrm{\πk}}{N}\right)))---\left(9\right)$

$F_{-1}=\frac{1}{N}(\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}(x_k\cos \left(\frac{2\mathrm{\πk}}{N}\right)-y_k\sin \left(\frac{2\mathrm{\πk}}{N}\right))+j\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}(y_k\cos \left(\frac{2\mathrm{\πk}}{N}\right)+x_k\sin \left(\frac{2\mathrm{\πk}}{N}\right)))---\left(10\right)$

N is the number of pixel on the maximum colour of skin UNICOM region contour, and k is the integer of 0...N-1, (x _k, y _k) be the coordinate of putting on the maximum colour of skin UNICOM region contour, j is the (j of imaginary unit ²=-1), F ₁, F _-1Represented respectively with counterclockwise and clockwise direction calculate the first order component of the Fourier transform of maximum colour of skin UNICOM region contour.Represent that when R=0 maximum colour of skin UNICOM zone is a border circular areas; Represent that when R=1 maximum colour of skin UNICOM zone is straight line; Represent that when R is between 0 and 1 maximum colour of skin UNICOM zone is an ellipse.

The degree of irregularity in maximum colour of skin UNICOM zone;

The degree of irregularity IR in maximum colour of skin UNICOM zone has represented the degree of region contour stray circle, adopts following formula to calculate the degree of irregularity IR in maximum colour of skin UNICOM zone:

$\mathrm{IR}=1.0-\frac{{\left|\right|F_1\left|\right|}^2+{\left|\right|F_{-1}\left|\right|}^2}{{\mathrm{\σ}}^2}---\left(11\right)$

Wherein, σ ²Represented to be provided the variance of having a few on the maximum colour of skin UNICOM region contour by following formula:

${\mathrm{\σ}}^2=\frac{1}{N}\underset{k=0}{\overset{N}{\mathrm{\Σ}}}(x_k^2+y_k^2)-({\left(\frac{1}{N}\underset{k=0}{\overset{N}{\mathrm{\Σ}}}{x}_k\right)}^2+{\left(\frac{1}{N}\underset{k=0}{\overset{N}{\mathrm{\Σ}}}{y}_k\right)}^2)---\left(12\right)$

The ratio of semi-minor axis length of the regional fitted ellipse of maximum colour of skin UNICOM;

The spindle tilt of the regional fitted ellipse of maximum colour of skin UNICOM;

Step 103: the evaluation of training model is also estimated the pornographic degree of input picture, the output evaluation result.

Evaluation model is selected the random forest sorter of Regression Model, and the random forest sorter is made of a series of decision trees, and in Regression Model, what random forests algorithm was exported is the mean value of all decision tree Output rusults.

At first the handmarking is carried out in the training image storehouse, normal picture mark 0.0 value, pornographic image mark 1.0 values.The image library sample good at mark extracts feature, and with these feature input evaluation models, to the evaluation model training, obtain model parameter, by the evaluation of training model can automatic learning go out pornographic image and normal picture said extracted to 31 dimensional feature space on characteristic distributions.The evaluation model that utilization trains is estimated input picture, and each decision tree in the random forest sorter all can independently be judged input picture, and export 0 or 1 the result of decision.Final Output rusults is the mean value of these results of decision, namely between 0%～100% evaluation result, has represented the pornographic degree of image, and numerical value is larger to show that the pornographic degree of image is higher, is convenient to pornographic image is carried out classified filtering.

The above; only be the embodiment among the present invention; but protection scope of the present invention is not limited to this; anyly be familiar with the people of this technology in the disclosed technical scope of the present invention; can understand conversion or the replacement expected; therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.

Claims (3)

1. Pornographic image evaluating method is characterized in that the method comprises:

Step 1: the colour of skin to input picture is cut apart, and obtains area of skin color;

Step 2: totally 31 dimensional feature constitutive characteristics are vectorial by global feature, body local genius loci and trunk shape facility three classes at the good great amount of images sample extraction input picture of handmarking, wherein, the global feature of described image comprises 14 features: the geometric moment of the colour of skin ratio of the entropy of image length breadth ratio, image, image, colour of skin connected region number and 10 dimension area of skin color; Described body local genius loci comprises 4 features: the ratio of the people's face quantity that detects, the chest quantity that detects, the private section quantity that detects and the shared integral image colour of skin of face complexion; Described trunk shape facility comprises 13 features: maximum colour of skin UNICOM region area accounts for the ratio of total colour of skin area, maximum colour of skin UNICOM regional center to the spindle tilt of the ratio of semi-minor axis length of the degree of irregularity in the circularity in the geometric invariant moment in the distance of picture centre, the maximum colour of skin of 7 dimensions UNICOM zone, maximum colour of skin UNICOM zone, maximum colour of skin UNICOM zone, maximum colour of skin UNICOM zone fitted ellipse and maximum colour of skin UNICOM zone fitted ellipse;

Wherein, the colour of skin ratio of described image is defined as:

The circularity R in described maximum colour of skin UNICOM zone calculates by following formula:

$R=\frac{{\left|\right|F_1\left|\right|}^2}{{\left|\right|F_{-1}\left|\right|}^2},$

Wherein, $F_1=\frac{1}{N}(\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}(x_k\cos \left(\frac{2\mathrm{\πk}}{N}\right)+y_k\sin \left(\frac{2\mathrm{\πk}}{N}\right))+j\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}(y_k\cos \left(\frac{2\mathrm{\πk}}{N}\right)-x_k\sin \left(\frac{2\mathrm{\πk}}{N}\right))),$

$F_{-1}=\frac{1}{N}(\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}(x_k\cos \left(\frac{2\mathrm{\πk}}{N}\right)-y_k\sin \left(\frac{2\mathrm{\πk}}{N}\right))+j\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}(y_k\cos \left(\frac{2\mathrm{\πk}}{N}\right)+x_k\sin \left(\frac{2\mathrm{\πk}}{N}\right))),$

N is the number of pixel on the maximum colour of skin UNICOM region contour, and k is the integer of 0...N-1, (x _k, y _k) be the coordinate of putting on the maximum colour of skin UNICOM region contour, j is imaginary unit, j ²=-1, F ₁, F _-1Represent respectively with counterclockwise and clockwise direction calculate the first order component of the Fourier transform of maximum colour of skin UNICOM region contour;

The degree of irregularity IR in described maximum colour of skin UNICOM zone represents the degree of region contour stray circle, adopts following formula to calculate:

$\mathrm{IR}=1.0-\frac{{\left|\right|F_1\left|\right|}^2+{\left|\right|F_{-1}\left|\right|}^2}{{\mathrm{\σ}}^2},$

Wherein, σ ²Represent the variance of having a few on the maximum colour of skin UNICOM region contour:

${\mathrm{\σ}}^2=\frac{1}{N}\underset{k=0}{\overset{N}{\mathrm{\Σ}}}(x_k^2+y_k^2)-({\left(\frac{1}{N}\underset{k=0}{\overset{N}{\mathrm{\Σ}}}{x}_k\right)}^2+{\left(\frac{1}{N}\underset{k=0}{\overset{N}{\mathrm{\Σ}}}{y}_k\right)}^2);$

Step 3: the evaluation of training model is also estimated the pornographic degree of input picture, output pornographic image evaluation result.

2. Pornographic image evaluating method according to claim 1 is characterized in that, described evaluation model is selected random forests algorithm, adopts the Regression Model of random forests algorithm.

3. Pornographic image evaluating method according to claim 1 is characterized in that, described evaluation result is the evaluation result of output 0%～100%, and numerical value is larger to show that the pornographic degree of image is higher, is convenient to pornographic image is carried out classified filtering.

Images