CN105068134B - A kind of method that dangerous material are concealed in utilization X-ray multi-view image detection footwear - Google Patents

Abstract

The invention relates to a method for detecting dangerous articles hidden in shoes by using X-ray multi-view images, the method comprising: using the multi-view images of the shoes to be inspected to determine the type of shoes to be inspected; looking for suspected dangerous articles according to the color component density detection model area; according to the geometric shape features, dynamic range of image grayscale and image material characteristics, the attributes of the suspected dangerous goods area are preliminarily judged; according to the texture feature classification model, the attributes of the suspected dangerous goods area with uncertain attributes after preliminary judgment are judged; using orthogonal In the angle-of-view electron density estimation method, attribute judgment is continued for suspected dangerous areas with uncertain attributes. By applying the method of the invention, it is possible to simply and quickly determine whether dangerous goods are hidden in the shoes, and to determine the hidden position and range of the dangerous goods. The invention can adapt to shoes of different sizes, shapes and materials, and can effectively detect whether it is a simple binding method or a hiding method of replacing the structure of the shoe itself.

Description

A Method of Using X-ray Multi-view Image to Detect Dangerous Articles Concealed in Shoes

technical field

The invention relates to the technical field of X-ray digital imaging safety inspection, in particular to a method for automatically detecting dangerous articles hidden in shoes by using X-ray multi-view images.

Background technique

At present, incidents of criminals hiding dangerous goods in luggage, personal clothes and shoes happen from time to time. For example, American Airlines discovered an attempt to detonate explosives hidden in shoes to create a terrorist incident, and Beijing Customs intercepted an incident of illegal elements hiding drugs in shoes at the Capital Airport. These terrorist incidents have brought many unrest factors to the society, which also makes the automatic inspection of dangerous goods in public places particularly important.

Published documents on safety inspection technology at home and abroad are mostly related to the identification of dangerous items in luggage. For example, US patents US20120093367A1, US20130003135A1, and its Canadian patent CA02608124 do not directly mention the identification and identification of dangerous items hidden in shoes. detection. The patent application No. 200910088495.7 applied by the First Research Institute of the Ministry of Public Security in China discloses a technology for detecting dangerous items in luggage by using dual-energy computing materials and multi-view reconstruction to obtain density. Dangerous goods; Another patent application number 201210424950.8 applied by the First Research Institute of the Ministry of Public Security discloses a method of using X-ray multi-view technology to check whether the liquid taken out of the luggage package and placed in a special inspection box is dangerous technology. So far, no automatic detection method directly related to hiding dangerous goods in shoes has been reported.

Contents of the invention

Aiming at the above-mentioned problems in the prior art, the present invention provides a method for automatically detecting explosives or drugs hidden in shoes by using X-ray multi-view security inspection equipment, which can quickly and effectively find explosives or drugs hidden in shoes by criminals.

In order to achieve the above object, the present invention adopts the following technical solutions.

A method for detecting dangerous goods hidden in shoes by using X-ray multi-view images, comprising the following steps:

Step 1. Use the multi-view images of the shoes to be inspected to determine the type of shoes to be inspected.

Step 2. Find the suspected dangerous goods area according to the color component density detection model.

Step 3: Preliminarily judge the attributes of the suspected dangerous goods area according to the geometric shape characteristics of the suspected dangerous goods area obtained in step 2, the dynamic range of the image grayscale, and the image material characteristics. The attributes include safety and danger. The attribute of the structural area of the shoe itself is safe, and the attribute of the area that replaces the structure of the shoe itself to hide dangerous goods is dangerous.

Step 4, according to the texture feature classification model, perform attribute judgment on the suspected dangerous goods area whose attribute is uncertain after the preliminary judgment in step 3.

Step 5, using the orthogonal viewing angle electron density estimation method, continue to judge the attributes of the suspected dangerous areas whose attributes are uncertain after the judgment in step 4.

Further, the multi-view images of the shoes to be inspected in step 1 are obtained by X-ray multi-view security inspection equipment by placing the shoes to be inspected in the inspection box.

Further, the step 1 also includes: performing grayscale morphology, threshold segmentation and binary morphology processing on the grayscale image of the multi-view image of the shoe to be inspected to segment the heel area from the entire shoe area.

Further, the shoes to be inspected are divided into four categories according to the way of hiding dangerous goods: shoes without heels and uppers, shoes without heels and uppers, shoes without uppers and heels, and shoes with uppers and heels; from the inner sole of the shoe to the edge of the upper Shoes with a height of no more than 70mm are without uppers, and those with a height of more than 70mm are uppers; shoes with a heel structure height of no more than 20mm are without heels, and those with a height greater than 20mm are with heels.

Further, the hiding method and position of dangerous goods in the four types of shoes are as follows: for heelless shoes, the dangerous goods are made into insoles and hidden on the soles of the feet; for heelless shoes, dangerous goods are made into flat sheets and hidden on the upper In the interlayer position, the shoes without heels hide dangerous goods in the hole formed by hollowing out the inner layer of the heel, and the shoes with heels hide dangerous goods in the hole in the heel and the interlayer of the upper.

Further, in step 2, the method for finding the suspected dangerous goods area according to the color component density detection model is as follows:

First, the Gaussian joint probability density function is used to calculate the probability p(X _i,j /Dan) that the pixel X _i,j is a dangerous product Dan in the color space, that is, the possibility, the formula is as follows:

In the formula, Dμ is the expected interval of dangerous goods Dan in the color space, and N is the number of pixels in the interval;

If p(X _i,j /Dan) is greater than the threshold, mark the pixel X _i,j as a pixel in the suspected dangerous area;

Then, use the spatial neighborhood information to calculate the similarity Com(R) of the pixel color vectors in the neighborhood R for the pixels marked as suspected dangerous areas:

In the formula, XC _{i, j} is the center element of the neighborhood, YC _{i, j} are other elements of the neighborhood, h is the evaluation of the complexity of the image; if Com(R) is greater than the threshold, the neighborhood is a suspected dangerous goods area; if If it is less than the threshold, then the pixel density of dangerous goods in the neighborhood is low, and the label of the suspected dangerous area is removed.

Further, the method for initially judging the attributes of the suspected dangerous goods area in step 3 is as follows:

The geometric shape characteristics of the suspected dangerous goods area include: area, geometric center, upper, lower, left, and right boundaries of the connected area, edge roughness, and angle between horizontal axes.

Further, the method for judging the attribute of the suspected dangerous area according to the texture feature classification model described in step 4 is as follows:

The gray level co-occurrence matrix is obtained by texture statistical analysis method. The gray level co-occurrence matrix describes texture features through spatial correlation characteristics. The correlation function p(g ₁ , g ₂ ) is:

In the formula, [(x ₁ ,y ₁ ),(x ₂ ,y ₂ )] represents a pixel pair composed of pixels (x ₁ ,y ₁ ), (x ₂ ,y ₂ ), f(x ₁ ,y ₁ ), f(x ₂ , y ₂ ) represent the gray levels of pixels (x ₁ , y ₁ ), (x ₂ , y ₂ ) respectively _, S represents the set of pixel pairs, S' represents the number of pixel pairs in the set S, The numerator on the right side of the equal sign represents the number of pixel pairs whose gray values are g ₁ and g ₂ respectively;

Based on the co-occurrence matrix, the features of the following 4 classifications are calculated:

In the formula, W _P represents energy; Cor represents correlation, μ _x and σ _x are the expectation and mean square error of the horizontal axis in p(g ₁ , g ₂ ), respectively, μ _y and σ _y are p(g ₁ , g 2 ), respectively The expectation and the mean square error in the vertical axis direction in g ₂ ); W _c represents the inverse gap; W _E represents the entropy; Borrowing the idea of Adaboost classification algorithm, according to the correct classification and misclassification performance of these four features in the training sample set, the energy is used as The first weak classifier, followed by entropy, correlation and inverse gap, combine the four weak classifiers to form a strong classifier; use the trained strong classifier to divide the suspected dangerous areas obtained in step 3 into three categories: Dangerous area; the structure of the shoe itself, that is, the safe area; uncertain suspected dangerous area.

Further, in step 5, the method for continuing to judge attributes by adopting the orthogonal view electron density estimation method is as follows:

Calculate the electron density of the pixel points in the uncertain suspected dangerous area, and judge the attribute of the area according to the difference between the electron density of the dangerous goods and the shoes to be inspected;

The electron density is equal to the ratio of the gray level of a pixel point in the image to the spatial distance passed by the rays forming the gray level of the point.

Further, after step 5, it also includes drawing a frame alarm for the dangerous attribute area.

Compared with the prior art, the present invention has the following advantages:

By applying the detection method of the present invention, it is possible to simply and quickly judge whether dangerous goods are hidden in the shoes, and determine the hidden position and range of the dangerous goods. The detection method of the invention is suitable for shoes of different sizes, shapes and materials. Regardless of the simple binding method or the hiding method of replacing the structure of the shoe itself, the method of the present invention can effectively detect.

Description of drawings

Fig. 1 is the schematic diagram of the special inspection box involved in the embodiment;

Fig. 2 is the overall flowchart of the method of the present invention;

Figure 3 is a schematic diagram of the estimated electron density of the orthogonal viewing angle: (a) shows the layout of the orthogonal viewing angle and the basic shape of the object to be measured in the channel, and (b) is a partial enlarged view of (a), showing how to calculate the distance passed by the ray.

detailed description

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

The schematic diagram of the special inspection box involved in the embodiment is shown in Figure 1. The bottom size of the plastic inspection box is 36cm*40cm. A foam pad is installed at the bottom of the inspection box. Hollow shoe print positions are made on the foam pad, and there is a gap between the two shoe prints. Put the two shoes side by side in the hollow shoe print of the inspection box, and use X-ray multi-view security inspection equipment to obtain multi-view images of the shoes.

A method for detecting dangerous goods hidden in shoes using X-ray multi-view images, as shown in Figure 2, comprises the following steps:

S1. Determine the type of shoe to be inspected according to the multi-view image of the shoe

According to the structure of the shoes and the means of simulating criminals hiding dangerous goods, the present invention divides the shoes into four categories, namely: shoes without a heel and no upper, shoes without a heel with a upper, shoes without a upper with a heel, and shoes with a upper and a heel. Shoes with a height of no more than 70mm from the inner bottom of the shoe to the edge of the upper are referred to as upperless shoes, and those with a height greater than 70mm are referred to as uppers;

According to the hiding space and concealment in each category, the present invention simulates lawbreakers and designs hiding means, among which: non-heeled shoes make dangerous goods into insoles and hide them on the soles of the feet, and non-heeled shoes make dangerous goods into flat pieces Hiding in the midlayer of the upper, the inner layer of the heel is hollowed out to hide dangerous goods in the heelless shoes, and the dangerous goods are hidden in the heel hole and the upper interlayer of the heeled shoes. The main method of classification is to perform grayscale morphology, threshold segmentation, and binary morphology processing on the grayscale image of the multi-view image to obtain the segmented area of the heel part and the whole shoe. According to the shoe structure research, the pre-defined size is used to judge whether there is an effective heel and an effective upper in the shoe to be inspected, so as to complete the classification. Steps S2-S5 in Fig. 1 are all carried out on the basis of the above-mentioned shoe classification, and steps S2-S5 are called in different ways, order and parameters in different classifications.

S2. Find the suspected dangerous goods area according to the color component density detection model.

All kinds of dangerous goods are substances with relatively fixed materials, and the material space is mapped to the color space, and various dangerous goods have their own characteristic areas in the color space. The color component density model uses the Gaussian joint probability density function to calculate the possibility that the pixel is a certain type of dangerous goods in the color space, and then uses the spatial information to jointly calculate the possibility of the area being a certain type of dangerous goods, and then according to the area is a certain type of dangerous goods. The ratio of the number of pixels with a high possibility of dangerous goods to the total number of pixels in the area, check the density of this possibility, and finally judge whether the area is a suspected dangerous area. Assuming that a certain type of dangerous goods is Dan, its expected interval in the color space is Dμ, and there are N pixels in the interval, then the possibility value p(X _{i ,j} /Dan )for:

where Con is the covariance matrix.

Formula (1) also embodies the concept of Mahalanobis distance, and the size of the distance also shows the similarity between the pixel and the dangerous goods. If the color of the pixel is far from the color of the dangerous goods, then the probability value p(X _i,j /Dan) will be low; on the contrary, the value of p(X _i,j /Dan) will be high. Set the prior threshold according to a large number of trainings. If p(X _{i, j} /Dan) is greater than the threshold, mark the pixel as a pixel in the suspected dangerous area. Then, the pixels marked as suspected dangerous areas are confirmed again by using the spatial neighborhood information. If there are no other pixels in the neighborhood that are marked as suspected dangerous areas, it is considered that the dangerous density is low, and the review is excluded, and the label of the suspected dangerous area pixels is removed; if there are other suspected dangerous area pixels in the neighborhood, the color of the pixels in the neighborhood R is calculated according to the following formula Vector similarity Com(R):

Among them, XC _{i, j} is the center element of the neighborhood, YC _{i, j} is the other elements of the neighborhood, h is the evaluation of the complexity of the image, the higher the complexity, the greater the value of h, and the higher the value of Com(R), indicating that the neighborhood The higher the domain pixel similarity, the higher the possibility of dangerous goods. Set the prior threshold according to a large number of trainings. If Com(R) is greater than the threshold, the neighborhood is a suspected dangerous area; if it is smaller than the threshold, then the pixel density of dangerous goods in the neighborhood is low, and the suspected dangerous area label is removed. The suspected dangerous area is represented by a binary value, and after binary morphological processing, it is used for other feature judgments below.

S3. Establish a set of geometric shape features of the suspected dangerous area, and comprehensively judge the attributes of the suspected dangerous goods area obtained in step S2 according to the geometric shape features of the suspected dangerous area, image grayscale and material features.

The attributes include safety and danger: the attribute of the structure of the shoe itself is safe, and the attribute of the dangerous goods that replace the structure of the shoe itself is dangerous.

The set of geometric shape features is a set of geometric statistical information of the connected area in the image, including: area, geometric center, upper, lower, left, and right boundaries of the connected area, edge roughness, and angle between horizontal axes. Because according to the classification of shoes, the amount of dangerous goods that can be hidden in different hiding positions is roughly fixed. This feature can be reflected by the area of the connected area of the transmission image. The geometric center, the upper, lower, left, and right boundaries of the connected area are the same. Due to the difference between the material and the structure of the shoe itself, dangerous goods generally have a relatively obvious boundary with the structure of the shoe itself, and the edge of the connected area will be relatively smooth. Use the roughness of the edge to judge whether the undetermined connected area is a dangerous area or the structure of the shoe itself. There is a class of dangerous goods that will be hidden on the sole of the shoe as an insole. The angle between the connected area of the image of such dangerous goods and the horizontal axis is within a certain range. This feature is used to judge the existence of such dangerous goods. Also based on shoe classification, the grayscale dynamic range of dangerous goods hidden in different locations is basically fixed. Check the grayscale range of the connected area and calculate the material value of the connected area to determine whether the area is an attribute of a suspected dangerous area. Among them, the final hazard attribute of the suspected dangerous goods hidden in the upper midlayer can be given through the judgment of step S3, and no further judgment is required; if the suspected dangerous goods hidden in the heel or made of insoles cannot be ruled out by the rules of step S3 Attributes need to be further judged according to the methods of steps S4 and S5.

S4. According to the texture feature classification model, perform attribute judgment on the suspected dangerous goods area whose attribute is uncertain after the preliminary judgment in step S3.

Generally, the heel or sole has reinforcing ribs, or a concave-convex structure for anti-slip, or other heel material structures are added to the hollow heel, and these structures will be reflected in the image texture features. On the contrary, dangerous goods are generally relatively uniform, with very inconspicuous texture features, or coarse texture. The invention adopts the texture statistical analysis method to obtain the gray level co-occurrence matrix. The gray level co-occurrence matrix describes texture features through spatial correlation characteristics, and its correlation function p(g ₁ , g ₂ ) is:

In the formula, [(x ₁ ,y ₁ ),(x ₂ ,y ₂ )] represents a pixel pair composed of pixels (x ₁ ,y ₁ ) _and (x ₂ ,y ₂ ), f(x ₁ ,y ₁ ) _, f(x ₂ ,y ₂ ) represent the grayscale of pixels (x ₁ ,y ₁ ) _and (x ₂ ,y ₂ ) respectively _, S represents the set of pixel pairs, S ^' represents the number of pixel pairs in the set S, The numerator on the right side of the equal sign represents the number of pixel pairs whose gray values are g ₁ and g ₂ respectively.

Some features can be calculated based on the co-occurrence matrix, in conjunction with the characteristics of the texture of soles in the present invention and the linear irrelevance characteristics between the features, select the following 4 features as classification:

Among them, W _P represents energy; Cor represents correlation, where μ _x and σ _x are the expectation and mean square error in the horizontal axis direction of p(g ₁ , g ₂ ), respectively, μ _y and σ _y are p(g ₁ , g 2 ), respectively g ₂ ) the expectation and the mean square error in the direction of the vertical axis; W _c represents the inverse gap; W _E represents the entropy. The above four features are calculated using the samples constructed when studying the algorithm. It is impossible for each feature to completely and correctly classify all samples, that is, each feature can train a weak classifier. Borrowing the idea of Adaboost classification algorithm, according to the correct classification and misclassification performance of these four features in the training sample set, energy is used as the first weak classifier, followed by entropy, correlation and inverse gap, and the four weak classifiers are combined together to form a strong classifier. Use the trained strong classifier to classify the suspected dangerous area in step S3. There are three types of classification results. One is to determine the dangerous area, and wait for the frame to display an alarm; the other is to determine the structure of the shoe itself, and wait for the image to be displayed;

S5. Using the orthogonal view electron density estimation method to continue to judge the attributes of the suspected dangerous areas with uncertain attributes in step S4, and give a final attribute judgment result.

The ratio of the gray level of a pixel in the image to the space distance that the ray passes through to form the gray level of the point represents the electron density of the measured substance at the actual position corresponding to the pixel. Dangerous goods and shoes are different in this feature, so this method can be used to judge the attributes of suspected dangerous areas. This method is more suitable for the substances to be inspected that have a certain height in the height direction of the X-ray equipment, that is to say, it is more suitable for estimating the situation of dangerous goods hidden in the heel of high-heeled shoes.

The schematic diagram of estimating electron density from orthogonal viewing angles is shown in Fig. 3 . In Fig. 3(a), V2 is the label of multi-angle, V4 is the label of bottom illumination angle, the dotted line indicates the imaging range that the angle of view can cover, and the solid line starting from the angle of view V2 indicates a ray passing through the substance to be measured. In Figure 3(b), the line segment AB is the actual ray passing through the object to be measured from multiple perspectives. The intersection points A and B of the angle ray and AB of the bottom view, through the known spatial arrangement of the detectors of the two angles of view and the positions of the detectors of the two points A and B in the image, the lengths of the line segment AC and the line segment BD are calculated, and the length of the line segment CD The length of can be obtained from the spatial arrangement of the bottom-illumination perspective detector and the position of the pixel detector on the image. The line segment EB is equal to the line segment CD, the length of the line segment AE is equal to the line segment AC minus the line segment BD, and the length of the line segment AB can be obtained in the right triangle AEB By the Pythagorean theorem and so on. In this way, the gray scale of the pixel point corresponding to the line segment AB on the image is longer than the length of the line segment AB, which can be expressed as the electron density of the pixel at this point. The premise of this calculation is assuming that the substance to be measured is a standard rectangle, but the heel is a trapezoid with a slightly longer upper side, which can be used as a method for estimating the electron density without affecting the accuracy. After the estimated electron density is obtained, it is finally judged whether the suspected dangerous area that does not give a result in step S4 is dangerous or not.

According to the results of step S3, step S4 and step S5, the image is finally displayed, and the dangerous area is marked with a red frame.

The present invention is not limited to the above-mentioned embodiments, and any obvious improvements or changes made by those skilled in the art to the above-mentioned embodiments will not exceed the concept of the present invention and the scope of protection of the appended claims.

Claims (5)

1. the method for dangerous material is concealed in a kind of utilization X-ray multi-view image detection footwear, it is characterised in that comprise the following steps：

Step 1, the type of footwear to be checked is determined using the multi-view image of footwear to be checked；

Step 2, doubtful dangerous material region is found according to color component Density Detection model；

First, pixel X is calculated using Gauss joint probability density function_i,jIt is dangerous material Dan Probability p in color space (X_i,j/ Dan), it is possible to property, formula is as follows：

<mrow> <mi>p</mi> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>/</mo> <mi>D</mi> <mi>a</mi> <mi>n</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <munder> <mi>&amp;Sigma;</mi> <mrow> <mi>n</mi> <mo>=</mo> <mn>1</mn> <mo>~</mo> <mi>N</mi> </mrow> </munder> <msup> <mrow> <mo>|</mo> <mrow> <mi>C</mi> <mi>o</mi> <mi>n</mi> </mrow> <mo>|</mo> </mrow> <mrow> <mn>1</mn> <mo>/</mo> <mn>2</mn> </mrow> </msup> </mrow> </mfrac> <mi>exp</mi> <mo>{</mo> <munder> <mi>&amp;Sigma;</mi> <mrow> <mi>n</mi> <mo>=</mo> <mn>1</mn> <mo>~</mo> <mi>N</mi> </mrow> </munder> <mrow> <mo>(</mo> <mo>-</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mrow> <mo>(</mo> <mrow> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>D&amp;mu;</mi> <mi>n</mi> </msub> </mrow> <mo>)</mo> </mrow> <mi>T</mi> </msup> <msup> <mi>Con</mi> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mo>(</mo> <mrow> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>D&amp;mu;</mi> <mi>n</mi> </msub> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>}</mo> </mrow>

In formula, D μ are dangerous material Dan interval in the expectation of color space, and N is interval interior number of pixels；

If p (X_i,j/ Dan) it is more than threshold value, by pixel X_i,jIt is designated as pixel in doubtful danger zone；

Then, pixel color in neighborhood R is calculated as follows to the pixel utilization space neighborhood information for being denoted as doubtful danger zone The similitude Com (R) of vector：

<mrow> <mi>C</mi> <mi>o</mi> <mi>m</mi> <mrow> <mo>(</mo> <mi>R</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>exp</mi> <mo>{</mo> <mo>-</mo> <mfrac> <mrow> <munder> <mo>&amp;Sigma;</mo> <mi>R</mi> </munder> <mo>|</mo> <mo>|</mo> <msub> <mi>XC</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>YC</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> </mrow> <mi>h</mi> </mfrac> <mo>}</mo> </mrow>

In formula, XC_i,jFor centre of neighbourhood element, YC_i,jFor neighborhood other elements, h is the evaluation to image complexity；If Com (R) is more than threshold value, and the neighborhood is doubtful dangerous material region；If less than threshold value, then the neighborhood dangerous material picture element density It is low, remove doubtful danger zone label；

Step 3, geometric characteristic, the dynamic range of gradation of image in the doubtful dangerous material region obtained according to the step 2 And iconography feature tentatively judges the attribute in doubtful dangerous material region；The attribute includes safety and dangerous, this body structure of footwear The attribute in region is safety, and the attribute that substituted for the region that footwear this body structures conceals dangerous material is danger；

Step 4, according to textural characteristics disaggregated model, to the uncertain doubtful dangerous material region of attribute after step 3 tentatively judgement Carry out determined property；

Gray level co-occurrence matrixes are obtained using texture statistics analytic approach, it is special that gray level co-occurrence matrixes describe texture by spatial correlation characteristic Levy, its relevance function p (g₁,g₂) be：

In formula, [(x₁,y₁),(x₂,y₂)] represent by pixel (x₁,y₁)、(x₂,y₂) composition pixel pair, f (x₁,y₁)、f(x₂,y₂) Pixel (x is represented respectively₁,y₁)、(x₂,y₂) gray scale, S represents pixel to set, and S ' represents the number of pixel pair in set S, etc. The molecule on number the right represents gray value respectively g₁And g₂Pixel pair number；

The feature of following 4 classification is calculated based on co-occurrence matrix：

<mrow> <msub> <mi>W</mi> <mi>P</mi> </msub> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <msub> <mi>g</mi> <mn>1</mn> </msub> </munder> <munder> <mo>&amp;Sigma;</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> </munder> <msup> <mi>p</mi> <mn>2</mn> </msup> <mrow> <mo>(</mo> <msub> <mi>g</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> <mo>)</mo> </mrow> </mrow>

<mrow> <mi>C</mi> <mi>o</mi> <mi>r</mi> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <msub> <mi>&amp;sigma;</mi> <mi>x</mi> </msub> <msub> <mi>&amp;sigma;</mi> <mi>y</mi> </msub> </mrow> </mfrac> <mrow> <mo>(</mo> <munder> <mo>&amp;Sigma;</mo> <msub> <mi>g</mi> <mn>1</mn> </msub> </munder> <munder> <mo>&amp;Sigma;</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> </munder> <msub> <mi>g</mi> <mn>1</mn> </msub> <mo>*</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> <mo>*</mo> <mi>p</mi> <mo>(</mo> <mrow> <msub> <mi>g</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> </mrow> <mo>)</mo> <mo>-</mo> <msub> <mi>&amp;mu;</mi> <mi>x</mi> </msub> <msub> <mi>&amp;mu;</mi> <mi>y</mi> </msub> <mo>)</mo> </mrow> </mrow>

<mrow> <msub> <mi>W</mi> <mi>c</mi> </msub> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <msub> <mi>g</mi> <mn>1</mn> </msub> </munder> <munder> <mo>&amp;Sigma;</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> </munder> <mfrac> <mn>1</mn> <mrow> <mn>1</mn> <mo>+</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>g</mi> <mn>1</mn> </msub> <mo>-</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </mfrac> <mo>*</mo> <mi>p</mi> <mrow> <mo>(</mo> <msub> <mi>g</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> <mo>)</mo> </mrow> </mrow> 1

<mrow> <msub> <mi>W</mi> <mi>E</mi> </msub> <mo>=</mo> <mo>-</mo> <munder> <mo>&amp;Sigma;</mo> <msub> <mi>g</mi> <mn>1</mn> </msub> </munder> <munder> <mo>&amp;Sigma;</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> </munder> <mi>p</mi> <mrow> <mo>(</mo> <msub> <mi>g</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> <mo>)</mo> </mrow> <mo>*</mo> <mi>log</mi> <mi> </mi> <mi>p</mi> <mrow> <mo>(</mo> <msub> <mi>g</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>g</mi> <mn>2</mn> </msub> <mo>)</mo> </mrow> </mrow>

In formula, W_PRepresent energy；Cor represents correlation, μ_x、σ_xIt is p (g respectively₁,g₂) in the expectation of horizontal axis and square Difference, μ_y、σ_yIt is p (g respectively₁,g₂) in vertical axis expectation and mean square deviation；W_cRepresent unfavourable balance away from；W_ERepresent entropy；Borrow Adaboost sorting algorithm thoughts, the correct classification concentrated according to this 4 features in training sample and mistake classification are showed, by energy Amount, as the first Weak Classifier, is then entropy, correlation and unfavourable balance successively away from 4 Weak Classifiers are jointly formed into one Strong classifier；With train come strong classifier the doubtful danger zone that step 3 is obtained is divided into three classes：Danger zone；Footwear sheet Body structure, i.e. safety zone；Doubtful danger zone is not known；

Step 5, using orthogonal views electron density method of estimation, to the uncertain doubtful hazardous area of attribute after step 4 judgement Domain proceeds determined property；

The multi-view image of footwear to be checked described in step 1, is by the way that footwear to be checked are put in inspection box, regarded by the X-ray more What angle Security Inspection Equipments were obtained；

The step 1 also includes：The gray level image of the multi-view images of footwear to be checked is carried out gray scale morphology, Threshold segmentation and Binary morphology processing, heel area is split from whole footwear region；

The footwear to be checked are divided into 4 classes in the way of dangerous material are concealed：Without with upper-free shoe, without side footwear are followed by, no side has with footwear, There is side to have with footwear；Bottom is upper-free shoe to footwear of the upper edges highly no more than 70mm on the inside of from footwear, and what it is more than 70mm is to have side footwear； Footwear with structure height no more than 20mm are that the footwear more than 20mm are to have with footwear without with footwear.

2. concealing the method for dangerous material in utilization X-ray multi-view image detection footwear according to claim 1, its feature exists In, dangerous material in the 4 class footwear conceal method and position is：Concealed without dangerous material are made into shoe-pad with upper-free shoe in sole Dangerous material are made flat and concealed in upper of a shoe interlayer position, no side, which has, conceals dangerous material in footwear with footwear by position without side footwear are followed by With in the cavate standby hole of internal layer, being followed by side footwear and concealing dangerous material in heel in hole and upper of a shoe interlayer.

3. concealing the method for dangerous material in utilization X-ray multi-view image detection footwear according to claim 1, its feature exists In described rapid 3 tentatively judge that the method for doubtful hazardous area Domain Properties is as follows：

The geometric characteristic in the doubtful dangerous material region includes：Area, geometric center, connected region border, side up and down Edge degree of roughness, trunnion axis angle.

4. concealing the method for dangerous material in utilization X-ray multi-view image detection footwear according to claim 1, its feature exists In, described in step 5 using orthogonal views electron density method of estimation proceed determined property method it is as follows：

The electron density for not knowing doubtful danger zone pixel is calculated, electronics itself is close with footwear to be checked according to dangerous material The difference of degree judges the attribute in the region；

The space length that the ray that the electron density is equal to the gray scale of a pixel with the formation gray scale in image is passed through Ratio.

5. the method for dangerous material is concealed in the utilization X-ray multi-view image detection footwear according to claim 1 any one, Characterized in that, also including alarming to dangerous attribute region picture frame after the step 5.