Randomly Acquired Characteristics Based Automatic Shoeprint Verification

Shoeprint verification refers to the comparison of the known shoeprint and questionable shoeprint to determine whether they come from the same shoe,which is important for case investigation.Randomly acquired characteristics refer the damages randomly formed when the shoe outsole is trod or bumped against a hard object during the wearing process,such as holes,rifts,and scratches.In forensic practice,randomly acquired characteristics are the main basis of the shoeprint verification by criminal investigation experts.The existing shoeprint verification algorithms mainly rely on the image local features,which lack attention to detect randomly acquired characteristics and use them for the shoeprint verification.However,shoeprint verification with randomly acquired characteristics is more in line with expert experience and actual need.Therefore,this thesis proposes a randomly acquired characteristics based automatic shoeprint verification method.The main work is as follows:(1)Randomly acquired characteristics detection and boundary description methods are proposed.For hole-type randomly acquired characteristic,the local morphological characteristics and global appearance similarity are combined to select the maximally stable extremal regions to obtain the detection results of holes.After that,the region growing method is used to extract the boundary of hole-type randomly acquired characteristic.For rift-type randomly acquired characteristic,DBSCAN algorithm is first used to cluster the skeleton line endpoints to extract the areas to be identified,and then rifts are identified based on the geometric characteristics of connected components.After that,the boundary of the rift is extracted based on the reconstruction of the rift region.Experimental results show that the proposed methods can effectively detect the hole-type and rift-type randomly acquired characteristic,and obtain their real boundaries.(2)In order to use the uniqueness of the spatial relationship of randomly acquired characteristics,a spatial relationship-based decision method of shoeprint verification is proposed.The proposed method first describes the distance relationship and direction relationship of randomly acquired characteristics based on the metric matrix and Delaunay triangulation.After that,the similarities of the corresponding spatial relationships in two shoeprints are measured,and the decision results of shoeprint verification are obtained according to the similarities of the spatial relationships.Experimental results show that the proposed method can be effective to reflect the identity of two shoeprints.(3)In order to combine the different advantages of randomly acquired characteristics and their spatial relationships in shoeprint verification,a multi-hierarchy decision method for shoeprint verification is proposed.In the proposed method,the matching rate of randomly acquired characteristics is firstly calculated,and when the matching rate does not satisfy the predefined condition,the verification conclusion of the two shoeprints is directly made.Otherwise,the verification decision of shoeprints based on the distance relationship and direction relationship are fused by random forest algorithm at the score level,and finally the shoeprints are verified according to the score.Experiments on the shoeprint dataset show that the proposed algorithm is competitive with an accuracy rate of 97%,false rejection rate of2.8%,and false acceptance rate of 3.2%.