Relatively permanent pigmented or vascular skin marks (RPPVSM) for forensic identification

• Main Author: Arfika Nurhudatiana
• Other Authors: Kong Wai-Kin Adams
• Format: Theses and Dissertations
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/61766
• Institution: Nanyang Technological University

Performing criminal and victim identification on evidence images can be very challenging at times due to lack of biometric traits. Gunmen, terrorists, and violent riot protesters often cover their faces with masks or clothing, making face recognition impossible. Pedophiles in child sexual abuse images often hide or blur their faces and tattoos to avoid recognition. Though non-facial body parts are commonly observable in the evidence images of these cases, very limited research has been done to use non-facial skin information for criminal and victim identification. This thesis proposes a novel biometric trait named Relatively Permanent Pigmented or Vascular Skin Marks (RPPVSM), which is easily observable on the skin, for forensic identification. In this thesis, legal and operational aspects of RPPVSM for forensic identification are explored. It is required by law that a new science or methodology presented by expert witnesses in U.S. federal trials is scientifically valid and can be properly applied to the facts at issue. Although skin marks have been regularly used in forensic investigation, RPPVSM have not been scientifically studied for personal identification. This thesis aims to address some fundamental questions regarding RPPVSM as a biometric trait, such as how many RPPVSM are sufficient for identification and what their potential error rates are by studying the individuality of RPPVSM patterns in population. This thesis also aims to develop an automated RPPVSM identification system which automatically detects and matches RPPVSM patterns in color images. As preprocessing, a skin segmentation algorithm is developed. The skin segmentation algorithm consists of a clustering operation performed in the YCbCr and normalized RGB color spaces and a histogram analysis to determine the optimal number of clusters in each input image. The RPPVSM detection algorithm consists of preprocessing, RPPVSM candidate detection, and classification. In the preprocessing, blue channel is extracted from the RGB color space and contrast-enhanced. RPPVSM candidates of different sizes are then detected from the preprocessed image by a multi-scale Laplacian of Gaussian (LoG) filtering operation followed by a binary thresholding operation. The detected candidates are then classified as RPPVSM and non-RPPVSM based on contrast, shape, size, texture, and color features using SVM, neural network, and decision tree classifiers. To match RPPVSM patterns, a non-rigid point matching method is employed for registration. Two aligned RPPVSM are considered to be matched if their distance is within a tolerance distance threshold. Since many Asian subjects have only a few RPPVSM on their skin, their RPPVSM patterns are sometimes not unique enough for identification. A fusion system which integrates RPPVSM with vein patterns is proposed to overcome this problem. The proposed RPPVSM identification system and the fusion were evaluated on a total of 3,560 images of backs, chests, arms, and thighs collected from 400 subjects in varying pose, viewpoint, scale, and illumination conditions. The proposed RPPVSM detection algorithm achieved rank-1 and rank-10 identification accuracies of 76.79% and 88.97% respectively, much better than the comparison methods which achieved rank-1 and rank-10 accuracies below 51% and 79% respectively. The fusion improves unimodal vein identification by 1% to 10%, depending on the number of RPPVSM available on the skin. These results signify the potential of the proposed RPPVSM identification system and its fusion for forensic investigation. To the best of our knowledge, this is the first and most comprehensive research on non-facial skin mark patterns for criminal and victim identification.