CN103955940A - Method based on X-ray back scattering image and for detecting objects hidden in human body - Google Patents

Abstract

The invention discloses a method based on an X-ray back scattering image and for detecting objects hidden a human body, and relates to the technical filed of computer image processing. The method is characterized in that the method includes the following steps that X-ray back scattering images are collected, and preprocessing is conducted on the obtained images; the threshold value calculation method with the combination of the global threshold value and the dynamic threshold value is conducted on the preprocessed images, and gray scale discontinuity caused by the dynamic threshold value is processed by the adoption of a smoothing technique; image segmentation is conducted by the adoption of the obtained threshold values, the ambiguity evaluation index, the intra-regional uniformity evaluation index and the inter-regional contrast evaluation index are adopted as feedback, the dynamic threshold value with the best segmentation effect is obtained in a self-adaptive mode, and finally an ideal segmentation effect is obtained. The segmented hidden objects are extracted by the adoption of a connected region labeling algorithm on the basis of the above steps. The method has the advantages of being high in work efficiency and precise in detection result.

Description

A kind of detection method of the human body cache based on X ray backscatter images

Technical field

The present invention relates to computer image processing technology field, particularly relate to a kind of detection method of the human body cache based on X ray backscatter images.

Background technology

Along with the research of domestic and international detection, assessment and emergency plan for dangerous matter sources is more and more deep, thereby being used to safety check field, X ray back scattering imaging technology carries out the detection that human body carries cache.Wherein for the image of scan image, cutting apart is the basic link of X ray backscatter images human body detection.Its target is from X ray backward scattering scan image, cache to be extracted from human body.Complete effective human body cache auto Segmentation has vital role for feature extraction, Classification and Identification and the harmful grade differentiation etc. of later stage cache.The scan image that the said equipment obtains at present mainly relies on supervisory personnel to carry out artificial interpretation, dangerous material Automatic Measurement Technique achievement in research for this type of image is very few, in addition, during due to X ray back scattering imaging, be subject to the position of object on human body, the impact of the factors such as clothing materials, human posture and illumination, makes human body cache fast and accurately cut apart difficulty increase.Therefore realize the automatic detection of computing machine to human body hiding article in X ray backscatter images, and then identify all kinds of dangerous material, thereby avoid expending a large amount of manpowers and the time has very important Research Significance.

Summary of the invention

The technical problem to be solved in the present invention is: the detection method that a kind of human body cache based on X ray backscatter images is provided.

The technical scheme that the present invention takes for the technical matters existing in solution known technology is:

A detection method for human body cache based on X ray backscatter images, comprises the steps:

S101, collection X ray backscatter images, carry out pre-service to X ray backscatter images, obtains pretreatment image;

S102, pretreatment image is carried out to the threshold calculations that global threshold combines with dynamic threshold, and utilize the gray scale uncontinuity that smoothing technique causes dynamic threshold to process; Detailed process is:

S1021, utilize otsu algorithm to carry out global threshold calculating to pretreatment image, obtain image overall threshold value t1;

S1022, pretreatment image is divided into a series of subimages, and each frame subimage is carried out to the calculating of otsu local threshold, obtain the local threshold t2 of each frame subimage, subsequently according to the discussion of classifying of the inter-class variance under this local threshold: if inter-class variance is less than the threshold of threshold constant of user's self-defining, utilize the global threshold t1 obtaining in S1021, according to formula

T=(1-α) t1+ α t2 obtains new threshold value t; Wherein, α is inter-class variance; Otherwise: new threshold value t=t2;

S1023, new threshold value t is deposited in matrix M, matrix M is carried out to smoothing processing, be specially:

Each threshold value element that each new threshold value t in matrix M and its are around existed in 4 neighborhoods is weighted addition, and replaces new threshold value t with this; Wherein the weights in 4 fields are respectively 0.1; After completing, smooth operation obtains new threshold matrix M2;

S103, utilize above-mentioned obtained threshold value to carry out image to cut apart, and utilize these three evaluation indexes of blur level, intra-zone homogeneity and interregional contrast as feedback, obtain adaptively the best dynamic threshold of segmentation effect and complete image binaryzation; Detailed process is:

S1031, utilize the threshold value element in threshold matrix M2 to carry out binarization segmentation processing to pretreatment image;

S1032, utilize the segmentation result after blur level, intra-zone homogeneity and interregional contrast are processed above-mentioned binarization segmentation to carry out comprehensive evaluation;

S1033, repeating step S1022, S1023, S1031, S1032 successively, until obtain the highest situation of cutting apart of evaluation of estimate;

S1034, utilize the situation of cutting apart of above-mentioned acquisition to carry out the threshold calculations that global threshold combines with dynamic threshold, complete image binaryzation;

S104, obtained binary image is carried out to connected component labeling, utilize number of pixels this condition in a certain scope that object comprises as screening, to obtain the extraction of final cache.

Further: the pre-service in described step S101 is specially:

S1011, image gray processing

Adopt the rgb pixel value of DIB structure extraction X ray backscatter images, and three components are weighted on average with different weights, wherein red pixel weight is 30%, and green pixel is 59%, and blue pixel is 11%;

S1012, image denoising

Adopt gaussian filtering to carry out denoising to image, the concrete operations of gaussian filtering are: by each pixel in a template scan image, go to substitute the value of convolution central pixel point with the weighted mean gray-scale value of pixel in the definite neighborhood of template; In Gauss's template, near the position of the centre of neighbourhood, its weights are just higher; So arrange the meaning of weights to be image detail to carry out level and smooth time, can retaining more the overall gray distribution features of image; Above-mentioned template size is 3*3, and the numerical computational formulas that in template, coordinate is (x, y) is:

$G(x;y,\mathrm{\σ})=\frac{1}{{2\mathrm{\π\σ}}^2}{\exp }^{-(x^2+y^2)/{2\mathrm{\σ}}^2}$

Wherein: σ is scale parameter, for determining the level and smooth degree of filtering.Take template center as initial point, x, y is that each position of template is with respect to the coordinate of initial point; According to above formula, build Gauss's template, and image is carried out to denoising.

Further: described step S104 is specially:

S1041, obtained binary image is carried out to connected component labeling, obtain a plurality of connected regions;

S1042, calculate the pixel number that each connected region comprises;

S1043, when pixel number, cross and be greater than or less than user-defined threshold value, thereby think it can not is that cache filters out final cache

Advantage and good effect that the present invention has are:

The present invention realizes the automatic detection of computing machine to the cache that in X ray backscatter images, human body carries, allow staff free from heavy vision work, and improve the degree of accuracy of detection to improve safety check speed and quality, the method can reach following effect: solve existing overall OTSU algorithm and cannot take image local details into account and cause the not good problem of segmentation effect, utilize comprehensive evaluation to calculate adaptively dynamic threshold, thereby be combined to complete with global threshold, cut apart.Thereby and for dynamic threshold, according to actual conditions, carry out smoothly reaching improving partitioning algorithm and completing the detection of human body cache and know the real situation; Simultaneously the advantage of the method is when image target object and background gray scale is relatively more approaching or thereby background and object grey scale change all can take into full account local detail when larger and obtain satisfied segmentation result.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the detection method of the human body cache based on X ray backscatter images in the present invention;

Fig. 2 is the first preferential embodiment design sketch of the human body cache based on X ray backscatter images in the present invention;

Fig. 3 is the second preferential embodiment design sketch of the human body cache based on X ray backscatter images in the present invention.

Embodiment

For further understanding summary of the invention of the present invention, Characteristic, at this, exemplify following examples, and coordinate accompanying drawing to be described in detail as follows:

Refer to Fig. 1, a kind of detection method of the human body cache based on X ray backscatter images, has been calculated and has been cut apart task by dynamic threshold and global threshold adaptively, and detailed step is:

S101: gather X ray backscatter images, X ray backscatter images is carried out to pre-service, obtain pretreatment image;

1) image gray processing

This specific embodiment adopts the rgb pixel value of DIB structure extraction X ray backscatter images, and three components are weighted on average with different weights, and wherein red pixel weight is 30%, and green pixel is 59%, and blue pixel is 11%;

2) image denoising

This specific embodiment adopts gaussian filtering to carry out denoising to image.The concrete operations of gaussian filtering are: by each pixel in a template scan image, go to substitute the value of convolution central pixel point with the weighted mean gray-scale value of pixel in the definite neighborhood of template.The simple smooth that is different from image is processed, and in Gauss's template, near the position of the centre of neighbourhood, its weights are just higher.So arrange the meaning of weights to be image detail to carry out level and smooth time, can retaining more the overall gray distribution features of image.The template size that this specific embodiment adopts is 3*3, and the computing formula of template is:

$G(x;y,\mathrm{\σ})=\frac{1}{{2\mathrm{\π\σ}}^2}{\exp }^{-(x^2+y^2)/{2\mathrm{\σ}}^2}$

According to above formula, build Gauss's template, and image is carried out to denoising.

S102: the image after pre-service is carried out to the threshold calculations that global threshold combines with dynamic threshold, and utilize the gray scale uncontinuity that smoothing technique causes dynamic threshold to process:

1) global threshold calculates

In this process, utilize otsu algorithm to carry out global threshold calculating.The method is called again maximum variance between clusters.For image I (x, y), the segmentation threshold of prospect and background is denoted as t ₁, the ratio that the pixel number that belongs to prospect accounts for entire image is designated as ω ₀, its average gray μ ₀; The background pixel ratio that accounts for entire image of counting is ω ₁, its average gray is μ ₁, the overall average gray scale of image is designated as μ, and inter-class variance is designated as g.

The size of supposing image is M * N, and the number of pixels that in image, the gray-scale value of pixel is less than threshold value T is denoted as N0, and the number of pixels that pixel grey scale is greater than threshold value T is denoted as N1, has:

ω ₀＝N0/M×N (1)

ω ₁＝N1/M×N (2)

N0+N1＝M×N (3)

ω ₀+ω ₁＝1 (4)

μ＝ω ₀*μ ₀+ω ₁*μ ₁ (5)

g＝ω ₀(μ ₀-μ)^2+ω1(μ ₁-μ)^2 (6)

By formula (5) substitution formula (6), obtain equivalence formula:

G=ω ₀ω ₁(μ ₀-μ ₁) ^2 (7) this be inter-class variance

Adopt the method traveling through to obtain making the threshold value of inter-class variance g maximum, be required.

2) dynamic threshold calculates

Image is carried out to piecemeal and obtain a series of subimages, and each frame subimage is carried out to local threshold calculate local threshold t ₂in (each frame subimage adopt 1), otsu algorithm obtains threshold value), and according to discussions of classify of the inter-class variance under this threshold value: if inter-class variance σ is less than threshold value threshold, utilize 1) the global threshold t of middle acquisition ₁, according to t=(1-α) t ₁+ α t ₂principle is weighted, and tries to achieve the new threshold value t of each piece, otherwise new threshold value t=t ₂.

3) threshold smoothing

By 2) the new threshold value of each number of sub images of gained deposits in matrix M, and matrix M is carried out to smoothing processing: each threshold value element that each new threshold value element in M and its are around existed in 4 neighborhoods is weighted addition, and replaces original threshold value with this.Wherein the weights in 4 fields are respectively 0.1.After completing, smooth operation obtains new threshold matrix M2.

S103: utilize above-mentioned obtained threshold value to carry out image and cut apart, and utilize these three evaluation indexes of blur level, intra-zone homogeneity and interregional contrast as feedback, obtain adaptively the best dynamic threshold of segmentation effect and complete image binaryzation.

1) image segmentation evaluation index is calculated

Utilize blur level, intra-zone homogeneity and interregional contrast to carry out comprehensive evaluation to segmentation result corresponding to piecemeal situation in process 2.

Wherein blur level refers to image is transformed from a spatial domain to fuzzy quality territory, utilizes fuzzy tolerance to weigh FUZZY MAPPING in segmentation effect the present invention of image and adopts a kind of Nonlinear Mapping model to be:

X wherein _ijfor the gray-scale value that (i, j) in original image locates, U _ijfor this pixel is mapped to fuzzy quality territory institute from spatial domain

Corresponding fuzzy value; α is fuzzy factor, 0≤α≤1, t=X _ij/ X _max.

Nonlinear function

g(t)＝(1-e ^-t)/(1+e ^-t)

In addition, the computing formula of interregional contrast GLC and gradation uniformity UM is as follows:

$\mathrm{GLC}=\frac{|f_1-f_2|}{f_1+f_2}$

Wherein, f ₁and f ₂the average gray value that represents adjacent two regions

Gradation uniformity

$\mathrm{UM}=1-\frac{1}{C}{\mathrm{\Σ}}_i\left\{{\mathrm{\Σ}}_{(x,y)\∈R_i}{[f(x,y)-\frac{1}{A_i}{\mathrm{\Σ}}_{(x,y)\∈R_i}f(x,y)]}^2\right\}$

The value of UM is larger, and the homogeneity that sign is cut apart each intra-zone in figure is better, cuts apart quality higher.Wherein C is normalized factor, R _ibe i piece cut zone, f (x, y) is the gray-scale value of pixel (x, y), A _iregion R _ipixel count.

2) iterative computation block count

Repeat in 102 2)-3) and process three in 1) step, obtain the highest piecemeal situation of evaluation of estimate.The grade being applied in embodiment of the present invention method is the known technology in data processing method, and the embodiment of the present invention does not repeat at this.

S104: obtained binary image is carried out to connected component labeling, utilize number of pixels this condition in a certain scope that object comprises as screening, to obtain the extraction of final cache.

Refer to Fig. 2 and Fig. 3, wherein, Fig. 2 a and 3a are X ray backscatter images, have the binary map after the accessed Threshold segmentation of method in the present invention in Fig. 2 b and 3b are; Fig. 2 c and 3c are the image after cache extracts.

Above embodiments of the invention are had been described in detail, but described content is only preferred embodiment of the present invention, can not be considered to for limiting practical range of the present invention.All equalizations of doing according to the present patent application scope change and improve, within all should still belonging to patent covering scope of the present invention.

Claims (3)

1. a detection method for the human body cache based on X ray backscatter images, is characterized in that: comprise the steps:

S101, collection X ray backscatter images, carry out pre-service to X ray backscatter images, obtains pretreatment image;

S102, pretreatment image is carried out to the threshold calculations that global threshold combines with dynamic threshold, and utilize the gray scale uncontinuity that smoothing technique causes dynamic threshold to process; Detailed process is:

S1021, utilize otsu algorithm to carry out global threshold calculating to pretreatment image, obtain image overall threshold value t1;

S1022, pretreatment image is divided into a series of subimages, and each frame subimage is carried out to the calculating of otsu local threshold, obtain the local threshold t2 of each frame subimage, subsequently according to the discussion of classifying of the inter-class variance under this local threshold: if inter-class variance is less than the threshold of threshold constant of user's self-defining, utilize the global threshold t1 obtaining in S1021, according to formula

T=(1-α) t1+ α t2 obtains new threshold value t; Wherein, α is inter-class variance; Otherwise: new threshold value t=t2;

S1023, new threshold value t is deposited in matrix M, matrix M is carried out to smoothing processing, be specially:

Each threshold value element that each new threshold value t in matrix M and its are around existed in 4 neighborhoods is weighted addition, and replaces new threshold value t with this; Wherein the weights in 4 fields are respectively 0.1; After completing, smooth operation obtains new threshold matrix M2;

S103, utilize above-mentioned obtained threshold value to carry out image to cut apart, and utilize these three evaluation indexes of blur level, intra-zone homogeneity and interregional contrast as feedback, obtain adaptively the best dynamic threshold of segmentation effect and complete image binaryzation; Detailed process is:

S1031, utilize the threshold value element in threshold matrix M2 to carry out binarization segmentation processing to pretreatment image;

S1032, utilize the segmentation result after blur level, intra-zone homogeneity and interregional contrast are processed above-mentioned binarization segmentation to carry out comprehensive evaluation;

S1033, repeating step S1022, S1023, S1031, S1032 successively, until obtain the highest situation of cutting apart of evaluation of estimate;

S1034, utilize the situation of cutting apart of above-mentioned acquisition to carry out the threshold calculations that global threshold combines with dynamic threshold, complete image binaryzation;

S104, obtained binary image is carried out to connected component labeling, utilize number of pixels this condition in a certain scope that object comprises as screening, to obtain the extraction of final cache.

2. detection method according to claim 1, is characterized in that: the pre-service in described step S101 is specially:

S1011, image gray processing

Adopt the rgb pixel value of DIB structure extraction X ray backscatter images, and three components are weighted on average with different weights, wherein red pixel weight is 30%, and green pixel is 59%, and blue pixel is 11%;

S1012, image denoising

Adopt gaussian filtering to carry out denoising to image, the concrete operations of gaussian filtering are: by each pixel in a template scan image, go to substitute the value of convolution central pixel point with the weighted mean gray-scale value of pixel in the definite neighborhood of template; In Gauss's template, near the position of the centre of neighbourhood, its weights are just higher; So arrange the meaning of weights to be image detail to carry out level and smooth time, can retaining more the overall gray distribution features of image; Above-mentioned template size is 3*3, and the numerical computational formulas that in template, coordinate is (x, y) is:

$G(x;y,\mathrm{\σ})=\frac{1}{{2\mathrm{\π\σ}}^2}{\exp }^{-(x^2+y^2)/{2\mathrm{\σ}}^2}$

Wherein: σ is scale parameter, for determining the level and smooth degree of filtering.Take template center as initial point, x, y is that each position of template is with respect to the coordinate of initial point; According to above formula, build Gauss's template, and image is carried out to denoising.

3. detection method according to claim 1 and 2, is characterized in that: described step S104 is specially:

S1041, obtained binary image is carried out to connected component labeling, obtain a plurality of connected regions;

S1042, calculate the pixel number that each connected region comprises;

S1043, the pixel number comprising when connected region are greater than or less than user-defined threshold value, thereby think it can not is that cache filters out final cache.