RU2595181C2 - Method for selecting images for identification of weapon from striker trace - Google Patents

Abstract

FIELD: weapons and ammunition.

SUBSTANCE: invention relates to identification of firearms from traces of striker with individual feature spot of arbitrary shape by processing digital images of traces of strikers and their further analysis. Investigated shell is scanned to produce original digital image of investigated striker trace in grayscale. Local variations of gradations are smoothed. Smoothed initial digital image of striker trace is converted into binary black-and-white image, in which background has one colour, and individual attribute of other. Individual feature descriptors are measured, which do not depend on orientation of image, represented by area S perimeter P, maximum I_max and minimum I_min moments of inertia. Method includes selecting similar images of striker trace with individual feature in form of spot of arbitrary shape from database of trace images. From traces selected from image database, sampling images of traces with closest values of descriptors S, P, I_max, I_min. Method includes presenting a loop of individual attribute of analysed trace and traces, selected from trace image database in form of N standard complex vectors. Finding magnitude of normalised scalar product of loop of analysed trace with loops selected from a database of traces. For identification of firearm, selecting from a database of images of traces with greatest value of maximum magnitude of normalised scalar product of loops.

EFFECT: higher efficiency of formation of priority of a list of images based on degree of similarity thereof with analysed trace by eliminating effect of orientation of images on comparison result, as well as accelerated selection of images based on degree of similarity thereof with the analysed by simplifying calculations.

2 cl, 7 dwg, 9 tbl

Description

The invention relates to the identification of firearms in the footsteps of strikers by processing digital images of traces of strikers and their subsequent analysis.

There is a method for selecting digital images of traces of strikers (Jie Zhou, Fanglin Chen, Jinwei Gu, A Novel Algorithm for Detecting Singular Points from Fingerprint Images, Pattern Analysis and Machine Intelligence, IEEE Transactions on Vol.31, No. 7, July 2009 pp. 1239 - 1250 .; Krestinin I.A., Seredin O.S. The method of singular points in the search for faces on graphic images // Izvestiya TulGU. Engineering. 2008. No. 3., 265 c.), Based on image recognition by special points, by highlighting special points according to one of the rules (inflection points, maximum gradients, lines and angles, etc.). Such extraction is carried out using various detectors, for example, a simplified Harris-Laplace detector, Sobel detector, etc. Next, they compare the special points on the test and test images by finding the cross-correlation function (PCF) or calculating various estimates.

The disadvantages of this method include the fact that not all descriptors are invariant to rotation, zooming, and illumination of the image. Additional difficulties are created by the variability of the striker footprint images received from the same weapon instance. As a result, when working on large arrays of digital images, it is impossible to correctly form a priority list according to the degree of coincidence of the studied images for identifying weapons.

There is a method of selecting digital images of tracks from strikers by finding the correlation function, based on determining the maximum or minimum of the FVC of two images by moving one relative to the other along the axes oX and oY, and rotating one relative to the other (Theodore, V. Vorburger, James, H .Yen, B., Bachrach, Thomas, B. Renegar, Li Ma, Hyug-Gyo Rhee, Xiaoyu, A. Zheng, Jun-Feng, Song; Charles, D. Foreman, Surface topography analysis for a feasibility assessment of a National Ballistics Imaging Database, NIST Interagency / Internal Report (NISTIR). 2007).

The disadvantages of this method include the need for time-consuming calculations, a long time to conduct a search on large data arrays, low efficiency due to the presence of various microinhomogeneities on the surface of the capsule associated with their production. As a result, the method does not allow to correctly form a priority list for the identification of weapons.

A known method for selecting digital images of tracks from strikers, based on the calculation of the proximity coefficient (Fabiano Riva, Christophe Champod, Automatic comparison and Evaluation of Impressions Left by a Firearm on Fired Cartridge Cases // J Forensic Sci, Vol. 59, No. 3, 2014, pp. 637-647). In this method, three numerical criteria are developed for assessing the similarity of images based on: Euclidean average distance between combined images; cross-correlation functions between coincident areas of the image; the average angle between the normal vectors to the surfaces of the compared images. Based on them, a proximity coefficient is introduced, which estimates the ratio of the probability that digital images of the strikers are received from the same weapon instance to the probability that the images are obtained for different weapons. For its calculation, a “joint” distribution is estimated, which is based on images from the same known striker trace and a “separate” distribution, for which known images of traces from deliberately different strikers are used.

The disadvantage of this method is that to build a joint and separate distribution (on the basis of which the similarity coefficient is estimated), it is necessary to operate with a large number of statistical data. The authors tested this method on 60 images of traces of one striker and on the same number of images of different weapons. In addition, at the initial stage, alignment of the studied images with respect to each other is required. The disadvantages should also include the dependence of the calculation results on the orientation of the compared images. Ultimately, this method does not correctly form a priority list for the identification of weapons.

Closest to the proposed one is a method for selecting digital images of tracks from strikers (Song J. Proposed “NIST Ballistics Identification System (NBIS)” Based on 3D Topographic Measurements on Correlation Cells // AFTE Journal, Volume 45 Number 2 - Spring 2013, pp. 184 -194), based on obtaining a digital image of the investigated track of the striker in grayscale, smoothing out local differences in brightness. Next, similar images of the trace are selected from the database of trace images; for this, 3D images of the test and test objects are formed, areas that are not suitable for identification are excluded from consideration, the images are divided into correlation cells, and FVCs are found for each cell. Identification is carried out if at least 6 cells in a row coincide (on these cells the maximum of the cross-correlation function is greater than a given level) or at least 3 cells in each of the two combinations (in total not less than 6).

However, constructing a cross-correlation function for 3-dimensional images is a rather time-consuming task, and searching through a large database of such images requires a lot of time. The calculation results depend on the orientation of the compared images, the recommendation list, formed according to the degree of coincidence of the images, allows for omissions (not falling into the top five of the recommended list of tracks that are in the test array and left in the same strikethrough as the track under study).

The technical result consists in increasing the efficiency of forming a priority list of images according to the degree of their similarity with the investigated trace by eliminating the influence of image orientation on the comparison result, accelerating the selection of images by the degree of their similarity to the studied by simplifying computational operations.

The technical result is achieved by highlighting the traces of the strikers with individual signs in the form of spots of arbitrary shape; measuring descriptors features that do not depend on the orientation of the image, such as area (S), the perimeter (P), the maximum and minimum moments of inertia (I _max; I _min); selection of images of traces with the closest values of descriptors S, P, I _max , I _min ; presenting the contour of an individual feature of the trail under study in the form of N standard complex-valued vectors, finding the module of the normalized scalar product of the contour of the trail under investigation with the contours of the selected traces; selection of the image with the maximum value of the maximum modulus of the normalized scalar product of contours for the identification of weapons.

If necessary, carry out an additional measurement on a binary image of the investigated trace striker Euler characteristics; a change in the binarization level of the image of the investigated striker trace in the direction of increasing or decreasing until the Euler characteristic changes; selection of images from the image database of traces according to new descriptor values.

The claimed invention is illustrated by drawings, where:

in FIG. 1 shows the trace of a striker with an individual sign in the form of a spot of arbitrary shape;

in FIG. 2 shows a striker trace with an individual attribute in the form of a set of arcs and circles;

in FIG. 3 shows the coordinate system of complex unit vectors;

in FIG. 4 shows an example of describing the boundary of an object by complex unit vectors, the boundary of the object is indicated by a dotted line;

in FIG. 5 shows the initial test pattern track pin represented in grayscale, and a binary image having Euler characteristic χ = 1 when binarization level Z = Z _0;

in FIG. Figure 6 shows the initial investigated image of the striker trace represented in gray gradations and its binary image having the Euler characteristic χ = 2 at the binarization level Z = Z ₀ ;

in FIG. Figure 7 shows the image of the striker trace obtained as a result of binarization by the Niblack method of the initial image under study at the level Z = Z ₀ -0.08.

The method is implemented as follows.

The test cartridges are scanned using a ballistic scanner or other digital technology with the possibility of macro photography traces of strikers. From the set of digital images of the tracks of the strikers located in the electronic database, images of the tracks with individual signs in the form of spots (areas) of arbitrary shape are selected. This allows us to exclude from the subsequent analysis images with individual characteristics with clearly different morphological characteristics. For example, it makes no sense to compare two tracks with each other, one of which has individual signs in the form of spots of arbitrary shape (Fig. 1), and the second as a set of circles and arcs (Fig. 2).

To reduce the negative impact of the roughness of the surface of the capsule and the traces of the mechanisms of its production, the initial images are smoothed by applying a median filter with a mask of at least 7 × 7 cells. Then, the smoothed image is converted to binary, where black is, for example, the background, and white is a sign. Adaptive binarization of images is carried out by the Niblack method with a mask of at least 25x25 cells. The result is a black and white image with a clearly defined border of individual characteristics. The use of adaptive binarization can significantly reduce the effect of uneven illumination of the striker trace on the selection of signs. The formation of binary images makes it possible to unambiguously and quickly determine such descriptors (characteristics) of attributes as their area, perimeter, etc.

Next, the following attribute descriptors are measured: S (area); P (perimeter); I _max (maximum moment of inertia); I _min (minimum moment of inertia). Each object under study is assigned coordinates in the attribute space corresponding to the value of its descriptors (S, P, I _max , I _min ). The Euclidean distance from the test object to each test object is calculated according to the formula (1), and then the nearest objects are selected from the test array. From these objects a priority (recommendation) list is formed according to the rule: the smaller the distance d _j , the closer to the top of the list the test object is located (the more it is similar in descriptors to the object under study). Comparison of images by descriptors that are independent of orientation and representing numerical values allows us to quickly and efficiently exclude from further analysis images that are very different from the trace being studied. This will significantly reduce the time for subsequent analysis of trace images.

, (1)

, (one)

where A and B are the test and test objects, respectively; i - i-th component of the feature vector (one of the descriptors: area, perimeter, etc.); j is the serial number of the object from the test array.

The contours of the traces of the studied trace and the traces of the priority list are described by N unit complex vectors that connect the points in accordance with the directions of the 8-connected system (Fig. 3). Unit vectors are assigned coordinates in the form of complex numbers: i; 1 + i; one; 1-i; -i; -1-i; -one; -1 + I (Furman, YA.A. et al. Introduction to contour analysis and its applications to image and signal processing // M. Fizmatlit. 2003. P. 25-32.). The maximum value of the modulus of the normalized scalar product | η | contours G and N, encoded in this way (Fig. 4), is invariant to their rotation, position on the plane and scale. Moreover, the module of the normalized scalar product (NSP) indicates the degree of similarity of the contours. Next, find the maximum modulus of the normalized scalar product | η | contours of signs of the test and test tracks of the striker according to the formula (2). The use of more than 70 unit complex vectors allows us to fairly accurately describe the boundaries of features, and therefore, to determine the degree of their similarity modulo NSP.

(2)

(2)

The final priority list is formed according to the rule: the higher the maximum value of the NSP module, the higher the image of the test object is in the priority list, the more the shape of the signs (the test and the test trace) are similar to each other. Representation of the boundaries of the compared features in the form of complex vectors allows you to quickly and easily automatically determine the degree of similarity of the shape of the characters regardless of the orientation of the images by using a simple operation - calculating the normalized scalar product module.

If the expert does not find in the first numbers of the priority list of pair traces, then the Euler characteristic is measured for the binary image of the studied trace of the striker, then the binarization level of the image of the investigated trace is changed in the direction of increasing or decreasing in steps of 0.01 until the Euler characteristic changes ( Fig. 5-7). After that, new values of the descriptors of the trait of the striker under investigation are determined. Then, according to the method described above, according to the new descriptor values and the new sign form, the studied trace is compared with the traces of the test array and a new priority list is formed. A change in the level of binarization until the Euler characteristic changes, allows one to take into account the probable variability of displaying on the image under study individual characteristics of the surface relief of the striker and, as a result, the variability of feature extraction during image binarization.

The operability of this method was tested on a test array of digital images of the tracks of strikers, including 210 objects. According to the results of the analysis of descriptors that are not dependent on the orientation of the images, a recommendation list of images was formed according to the degree of proximity of the descriptors. Using the first 20 images of the recommendation list with similar descriptor values, a pair of traces was searched for by the maximum of the normalized scalar product module and priority lists were generated. The results are presented in tables 1-9, which are formed from the first 15 numbers of priority lists. Traces are numbered as follows: the first number means a weapon instance, the number in brackets is the serial number of the cartridge case shot in this weapon instance. It can be seen that in all cases the pair trace was in the first three objects of the priority list with the exception of trace 9 (1), for which there is no pair trace in the database.

Table 1. The first 15 numbers of the priority list for the investigated trace 1 (1). Zero number is presented to control the correctness of calculations. N - serial number of the image in the priority list.

N Test. track Mach NSP 1 (1) 0 1 (1) one one 1 (2) 0.94 2 6 (1) 0.82 3 5 (2) 0.81 four 6 (2) 0.80 5 5 (1) 0.80 6 39 (1) 0.79 7 39 (2) 0.77 8 13 (1) 0.74 9 4 (1) 0.58 10 31 (1) 0.56 eleven 47 (1) 0.56 12 4 (2) 0.55 13 21 (2) 0.50 fourteen 21 (1) 0.45

Paired track 1 (2) in the first place of the priority list.

Table 2. The first 15 numbers of the priority list for the investigated trace 39 (1). Zero number is presented to control the correctness of calculations. N - serial number of the image in the priority list.

N Test. track Mach NSP 1 (1) 0 39 (1) one one 6 (1) 0.82 2 39 (2) 0.80 3 1 (1) 0.79 four 1 (2) 0.79 5 5 (1) 0.79 6 5 (2) 0.78 7 6 (2) 0.77 8 13 (1) 0.70 9 47 (1) 0.57 10 4 (1) 0.55 eleven 4 (2) 0.54 12 31 (1) 0.52 13 21 (2) 0.42 fourteen 1 (1) 0.38

Paired track 39 (2) in second place on the priority list.

Table 3. The first 15 numbers of the priority list for the investigated trace 5 (1). Zero number is presented to control the correctness of calculations. N - serial number of the image in the priority list.

N Test. track Mach NSP 5 (1) 0 5 (1) one one 6 (1) 0.828 2 5 (2) 0.823 3 1 (2) 0.820 four 1 (1) 0.80 5 6 (2) 0.80 6 39 (1) 0.79 7 39 (2) 0.77 8 13 (1) 0.74 9 4 (2) 0.62 10 4 (1) 0.58 eleven 47 (1) 0.56 12 31 (1) 0.55 13 21 (2) 0.43 fourteen 21 (1) 0.42

Pair track 5 (2) in second place on the priority list.

Table 4. The first 15 numbers of the priority list for the investigated trace 31 (1). Zero number is presented to control the correctness of calculations. N - serial number of the image in the priority list.

N Test. track Mach NSP 31 (2) 0 31 (2) one one 31 (1) 0.81 2 21 (1) 0.80 3 21 (2) 0.77 four 4 (1) 0.69 5 4 (2) 0.60 6 47 (1) 0.57 7 1 (1) 0.53 8 6 (2) 0.53 9 39 (2) 0.50 10 1 (2) 0.50 eleven 5 (2) 0.45 12 39 (1) 0.42 13 13 (1) 0.42 fourteen 5 (1) 0.41

Paired track 31 (1) in first place on the priority list.

Table 5. The first 15 numbers of the priority list for the investigated trace 6 (1). Zero number is presented to control the correctness of calculations. N - serial number of the image in the priority list.

N Test. track Mach NSP 6 (1) 0 6 (1) one one 5 (2) 0.85 2 6 (2) 0.84 3 5 (1) 0.82 four 39 (1) 0.82 5 1 (1) 0.82 6 1 (2) 0.80 7 39 (2) 0.78 8 13 (1) 0.68 9 4 (1) 0.58 10 47 (1) 0.57 eleven 4 (2) 0.55 12 31 (1) 0.52 13 21 (2) 0.43 fourteen 21 (1) 0.42

Pair track 6 (2) in second place on the priority list.

Table 6 The first 15 numbers of the priority list for the investigated trace 13 (2). Zero number is presented to control the correctness of calculations. N - serial number of the image in the priority list.

N Test. track Mach NSP 13 (2) 0 13 (2) one one 13 (1) 0.73 2 1 (2) 0.66 3 5 (2) 0.65 four 1 (1) 0.65 5 39 (1) 0.65 6 5 (1) 0.64 7 6 (2) 0.64 8 39 (2) 0.63 9 6 (1) 0.63 10 4 (2) 0.59 eleven 4 (1) 0.57 12 47 (1) 0.55 13 21 (2) 0.50 fourteen 31 (1) 0.47

Pair track 13 (1) in first place on the priority list.

Table 7. The first 15 numbers of the priority list for the investigated trace 4 (1). Zero number is presented to control the correctness of calculations. N - serial number of the image in the priority list.

N Test. track Mach NSP 4 (1) 0 4 (1) one one 4 (2) 0.72 2 47 (1) 0.61 3 5 (2) 0.59 four 5 (1) 0.58 5 1 (1) 0.58 6 1 (2) 0.57 7 13 (1) 0.56 8 39 (1) 0.55 9 39 (2) 0.53 10 31 (1) 0.53 eleven 6 (2) 0.53 12 6 (1) 0.51 13 21 (2) 0.51 fourteen 21 (1) 0.48

Pair track 4 (2) in the first place of the priority list.

Tab. 8. The first 15 numbers of the priority list for the investigated trace 47 (2). Zero number is presented to control the correctness of calculations. N - serial number of the image in the priority list.

N Test. track Mach NSP 47 (2) 0 47 (2) one one 47 (1) 0.73 2 31 (1) 0.62 3 4 (1) 0.59 four 6 (2) 0.58 5 21 (2) 0.58 6 21 (1) 0.56 7 5 (2) 0.53 8 4 (2) 0.52 9 39 (2) 0.52 10 6 (1) 0.51 eleven 5 (1) 0.50 12 39 (1) 0.50 13 1 (1) 0.50 fourteen 13 (1) 0.48

Paired track 47 (1) in first place on the priority list.

Tab. 9. The first 15 numbers of the priority list for the investigated trace 9 (1), which does not have a pair trace in the database. Zero number is presented to control the correctness of calculations. N - serial number of the image in the priority list.

N Test. track Mach NSP 9 (1) 0 9 (1) one one 21 (2) 0.66 2 31 (1) 0.56 3 21 (1) 0.54 four 47 (1) 0.44 5 4 (1) 0.42 6 4 (2) 0.35 7 6 (2) 0.35 8 13 (1) 0.32 9 1 (2) 0.31 10 5 (1) 0.28 eleven 1 (1) 0.27 12 39 (2) 0.27 13 6 (1) 0.26 fourteen 5 (2) 0.25

The rather low value of the maxima of the NSP module of the loops in Table 9 signals the possible absence of a pair of striker tracks in the image database. For the studied trace 9 (1), the pair trace is absent in the database.

The achieved technical result consists in increasing the efficiency of forming a priority list of images according to the degree of their similarity with the investigated trace by eliminating the influence of image orientation on the comparison result, accelerating the selection of images according to the degree of their similarity with the studied by simplifying computational operations.

In the test tests, we used a computer with the following characteristics: 1 gigabyte of RAM, 512 megabytes of video memory, Pentium E655 processor, GeForce 210 graphics card. The database included 210 similar images of striking marks with signs (individual features of the relief) in the form of spots of arbitrary shape. The search time in the automatic mode by the track under study in the database and the formation of the priority list did not exceed 5 minutes on average. It should be noted that the software was implemented in the Matlab 7.7.0 environment, and when switching to a programming language such as C ++, the search time can be reduced by several times.

findings

Test tests showed the efficiency of the proposed method, the effectiveness of its application for the correct selection of images according to the degree of their similarity with the investigated trace, as well as the short time required for automatic analysis of images throughout the database.

Claims (2)

1. A method for identifying weapons on the track of a striker with an individual sign in the form of a spot of arbitrary shape, including scanning the test sleeve to obtain an initial digital image of the investigated trace of the striker in gray gradations, smoothing out local gradients and selecting similar images of the trace of the striker with an individual sign in the form of a spot of arbitrary shape forms from the database of trace images, characterized in that the resulting smoothed original digital image of the striker trace is converted into a binary black and white image e, wherein the pattern has one color, and the individual attribute other measured descriptors individual feature not depending on the orientation of the image, which is used as S area, perimeter P, a maximum I _max and minimum I _min moments of inertia of the selected from a database of images traces traces carried sampled images closest descriptor values S, P, I _max, I _min, represent individual contour characteristic of the test trace and trace, selected from the database of images of tracks in a standard N kompleksnoznach s vectors are unit normalized scalar product of the test circuit trace the contours of the selected base tracks and arms to identify selected from the database image with the largest module tracks the maximum value of the normalized inner product circuit. 2. The method according to p. 1, characterized in that it additionally measures the Euler characteristic on the binary image of the investigated strike trace, changes the binarization level of the image of the studied strike trace to increase or decrease until the Euler characteristic changes, and then select images from the trace image database new descriptor values.

Images