Shoeprint Individual Characteristics Detection And Representation Algorithm

Shoeprint quantitative inspection plays a vital role in case detection.Shoeprint quantitative inspection gives the quantitative inspection conclusions of the two shoeprints which is mainly through the quantitative description of shoeprint individual characteristics and wear characteristics to analyze the specificity of features in two shoe prints.The individual characteristic refers to the characteristic caused by production or wearing that is different from the pattern,which can express the uniqueness of the shoe.And individual characteristics are also called random characteristics due to the randomness of the position and shape of the characteristics.However,the individual characteristics of shoeprint are all manually annotated by shoeprint experts at present.In view of this situation,the automatic detection and representation of the individual characteristics of shoeprints are necessary research.The detection and representation of individual characteristics can effectively avoid personal error caused by manual labeling,and can ensure the reliability of the shoeprint inspection algorithm when processing a large amount of footprint data quickly.Due to the individual characteristics’ s specificity and randomness and the complexity of the sole patterns,a single traditional detection algorithm cannot meet the detection needs.In order to improve the accuracy and efficiency of detection and accurately represent the boundaries of individual characteristics,this thesis proposes an automatic detection and representation algorithm for individual characteristics.The main work is as follows:1)Foreign object characteristics detection algorithm based on distance frequency factor is proposed.The foreign object characteristics usually appear as a blob area,which is obviously different in appearance from other individual characteristics.In order to better detect the foreign object characteristics,the blob feature area detection algorithm is used firstly,and then similarity screening for the differences in the distribution and morphology of the foreign object characteristic is performed,and the distance frequency factor to correct the similarity screening is used finally.In the screening,the directionality and roundness of the features in the feature area are compared to obtain an accurate foreign object characteristic area.This algorithm tests on the shoeprint image dataset annotated by shoeprint experts;the recall rate can reach 80.3%;the precision rate can reach 71.2%.2)Individual characteristics detection algorithm based on edge distance distribution features is proposed.Existing detection algorithms cannot adapt to complex and changeable individual characteristics in individual characteristic detection.In order to improve the accuracy of detection,this thesis proposes a individual characteristic detection algorithm based on edge distance distribution features.In addition,the difference in the distribution of different individual characteristics in the connected domain of the individual characteristic points is combined to improve the clustering algorithm in the cross-pattern clustering problem.The algorithm tests on the shoeprint image dataset annotated by shoeprint experts.The recall rate can reach 71.3%;the precision rate reaches 75.5%.3)Individual characteristics representation algorithm based on difference of Gaussian is proposed.The intermediate region methods cannot be applied to practice due to its limited accuracy and certain errors.In order to make boundary more accurately represent the detected individual characteristics,this thesis proposes a individual characteristics expression algorithm based on the difference of Gaussian.The algorithm improves the scattered edge of the intermediate region from black to white,and more in line with the actual boundary of the characteristics.In this thesis,shoeprint individual characteristics detection and representation algorithm is proposed.This algorithm tests on the shoeprint image dataset annotated by shoeprint experts.The recall rate of the overall individual feature detection can reach 72.2%;the precision rate can reach 75.1%.