A novel multiresolution-based hybrid approach for 3D footwear outsole feature classification and extraction

Footwear impressions retrieved from crime scenes are often due to shoes in various stages of wear. This forensic-related research on footwear recognition is able to extract information-rich 3D outsole patterns and produce 2D shoe-prints regardless of different degrees of wear. Based on pattern characteristics, outsoles are categorized into two types, Convex-Pattern-Dominant Outsoles (Convex-PDOs) and Concave-Pattern-Dominant Outsoles (Concave-PDOs). Initial work for extracting 3D Features from Concave-PDOs is reported in this paper. In our proposed method, outsole models are first captured using a 3D scanner. Patterns corresponding to higher and lower curvature variations are subsequently classified using a multiresolution-based curvature analysis approach. In a subsequent step to discard outliers from the extracted 3D features, by modifying contours of 3D outsole models, a pyramid method is employed to generate composite results. Visual analysis on current experimental investigations shows promising results for further 3D feature extraction and 2D shoeprint generation.