Registration Strategy For Ballistics Signatures Based On SURF Feature

In a narrow sense,ballistics signatures refer to the tool-marks on the surface of the bullets and cartridge cases which are picked up from the topographies of the gun parts.The signatures are a kind of microscopic geometric topographies,which are different from the macroscopic shape features of the images.They contain class features of guns such as shape and ripple,as well as gun individual information such as roughness,which reflect the surface structures and texture characteristics of the corresponding gun parts.Owing to their uniqueness,the ballistics signatures can be used for feature comparisons to identify the corresponding firearms.The traditional identification method of ballistics signatures is manual,which is based on visual observation and experience.This method is not only lack of objectivity,but also inefficient,which can not complete rapid and accurate identification of ballistics signatures.Therefore,the research of scientific and effective automatic identification system of ballistics signatures is of great significance to the whole field of bullet marks identification.In this paper,the microscopic geometric topographies of breech face impressions is taken as the research object.The feature extraction of the roughness and other individual information is carried out by using local invariant features.The automatic recognition and matching algorithm of breech face impressions is established,and the scientific and efficient recognition and registration of breech face impressions is realized.In this paper,theoretical analysis,experimental verification and other methods are used to conduct research on the registration strategy based on SURF feature.The main contents are summarized as follows:(1)According to the 3D microscopic geometric topographies of the breech face impressions,the collection method and the difficulty of identification are pointed out.Based on the characteristics of the ballistics signatures,SURF algorithm is used for feature extraction and description of the ballistics signatures.Scale space pyramids are constructed by using integral images and DoH approximation concepts to search and locate SURF feature points of breech face impressions.For the key-points located by SURF algorithm,the distribution of Haar wavelet response are used to describe the key-points,and the direction of key-points and the modulus of feature vectors are determined.Through the comparison between theories and the analysis of specific experiments,SURF algorithm has better comprehensive performance than SIFT algorithm.(2)The nearest neighbour ratio method based on Euclidean distance is used for the initial matching of ballistics signatures.On this basis,the relative distance is used to finish filtering and purification of the initial matching.Through the analysis of the results,the RANSAC algorithm is adopted to achieve further optimization,the recognition and matching between the breech face impressions are realized,and it is verified that the local invariant feature is scientific and adaptive to the description of the microscopic geometric topographiesof ballistics signatures.(3)With the help of the breech face impressions provided by the national institute of standards and technology(NIST),the relationship between the parameters of the SURF feature and matching results is analyzed.Through the observation and comparison of experimental data,the optimal parameter configuration and the identification conditions are confirmed,and the registration strategy of the breech face impressions is proposed based on SURF algorithm and RANSAC optimization.(4)On the basis of realizing the match and purification of breech face impressions,RANSAC algorithm has a certain randomness.Although the number of effective matching pairs after purification remains unchanged,the refining results are different.In order to overcome the randomness of RANSAC algorithm to support the objectivity of tool-marks identification,this paper introduces the consistent feature matching method based on support vector regression,and carries out an experimental test with the help of breech face impressions.Through the processing and analysis of test results,consistent matching of breech face impressions is realized,randomness is eliminated,and the optimization for the registration of ballistics signatures is completed.The effectiveness of SVR is verified,and the feasibility of SURF algorithm applied in the identification of ballistics signatures is confirmed.