Research On The Automatic Identification Method Of Cartridge Marks Impression Based On SIFT Feature

Gun-related cases are usually of bad nature and extremely harmful.They must be detected and stopped as soon as possible to ensure social stability and public safety.Bullet mark test is an important technical means for investigating gun-related cases,which can provide clues and evidence for solving cases for public security and judicial departments.The traditional bullet mark inspection method mainly relies on bullet mark inspectors to observe the details of bullet marks on the surface of the bullet through a microscope to achieve bullet comparison.The whole process is time-consuming,heavy workload and strong subjectivity.It urgently needs an automatic bullet mark identification technology to make up for the shortcomings of previous manual inspection.As an important material for bullet trace detection,the firingpin and breechface record the structure and texture characteristics of the surface of the traces,showing three-dimensional surface micro-geometry.The surface wavelength components can be divided into shape,waviness and roughness.From the experimental point of view,the feature points of rough surface are more manifested in convex peaks or valleys,and the actual identification work is basically centered on these more prominent features of peaks and valleys.The feature points extracted by SIFT are mostly speckle structures.These speckle structures are just the most basic forms of convex peaks or valleys.Therefore,SIFT has a natural adaptability to the description of complex texture morphology,and is very suitable for feature extraction and comparison of bullet marks such as firingpin and breechface.In order to propose a method for automatically identifying cartridge marks,this paper takes the firingpin and breechface marks as the research object,and uses SIFT,RANSAC combined algorithm and an empirical method to match two kinds of bullet trace samples.In this paper,SIFT algorithm is used to study the problem of bullet trace matching by theoretical analysis and comparative experiments.The main contents of this paper are as follows:(1)In order to explain the objective accuracy of SIFT algorithm in matching bullet traces,200 bullet shell samples were collected from the Center for Evidence Identification for experimental verification and trace pretreatment.The bullet mark data obtained by the microscope are processed,including pruning,filtering,enhancement,truncation and so on,and the gray image of 0-255 bullet shell trace is obtained.After comparing the advantages and disadvantages of different kinds of marks,the firingpin and breechface are selected as the following experimental samples.(2)To achieve rapid and accurate matching of bullet samples,a bullet trace registration method based on scale invariant feature transformation(SIFT)algorithm is prenested.In this paper,the implementation principle of SIFT algorithm is analyzed in detail and verified step bystep,including the construction of Gauss pyramid and Gauss difference pyramid to determine local extremum points;eliminating low contrast extremum points and edge response extremum points;generating feature description vectors and normalizing them.(3)In this paper,10 groups of firingpin and breechface were matched by experiments,and it was found that the matching logarithm of firingpin was less.The concept of homomorphic filtering is introduced to process 10 groups of firingpin samples with homomorphic filtering and then match the feature points.The matching effect is improved greatly.RANSAC algorithm is used to purify 10 groups of firingpin and breechface marks matching pairs.The results show that the matching effect of breechface marks is better,so the breechface marks are taken as the experimental object.Finally,the matching experiment is carried out again under the change of occlusion,size,rotation and other factors,which verifies that SIFT algorithm is strongly robust to these factors.(4)This paper introduces an empirical method of area proportion of dense area of feature points in detail.Firstly,we introduce the relative centralized feature points acquisition and feature point-intensive region extraction algorithm,with the minimum convex polygon surrounded by each feature point as the feature point-intensive region.Using gray histogram method to judge the similarity of the dense area of two breechfaces,it can be found that the cosine angle of the dense area of two breechfaces is generally between 4°and 6°through experimental comparison,the cosine angle of the unmatched area is more than 15°.Finally,the method of qualitative judgment of the matching of the two breechfaces is introduced,that is,the method of dense feature points.Based on the percentage of the dense area of feature points to the total area of breechface,the experiment proves that the percentage is more than 12%,then the two breechfaces are considered to be matched,and vice versa,they are unmatched.