Uncovering vein patterns from color skin images for personal identification in forensic investigation

• Main Author: Tang, Chaoying
• Other Authors: Kong Wai-Kin Adams
• Format: Theses and Dissertations
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition
• Online Access: https://hdl.handle.net/10356/51059
• Institution: Nanyang Technological University

Recent technological advances have allowed for a proliferation of digital evidence images. Using these images as evidence in legal cases (e.g. child sexual abuse, child pornography and masked gunmen) can be very challenging, because the faces of criminals or victims are not visible. Although large skin marks and tattoos have been used, they are ineffective in some legal cases, because the skin exposed in evidence images have neither unique tattoos nor enough skin marks for identification. The blood vessel between the skin and the muscle covering most parts of the human body is a powerful biometric trait, because of its universality, permanence and distinctiveness. Traditionally, it was impossible to use vein patterns for forensic identification, because they were not visible in color images. All the current vein recognition systems developed by companies and research laboratories rely on near infrared (NIR) imaging devices to capture high quality vein patterns from hand and wrist, where skin is relatively thin, for commercial applications. Up until now, no one studies vein patterns for forensic identification because no method has been developed to visualize vein patterns hidden in color images. The primary aim of this research is to develop algorithms for visualizing vein patterns hidden in color images so that criminal and victim identification can be performed based on vein patterns. The secondary aim of this research is to develop algorithms for removing JPEG blocking artifacts in skin images that adversely affect forensic recognition. We propose two approaches for uncovering vein patterns from color skin images. The first approach is based on RGB-NIR mapping. It extracts information from a pair of synchronized color and NIR images and uses a neural network (NN) to map RGB values to NIR intensities. Furthermore, we design an automatic intensity adjustment scheme for illumination compensation and an NN weight adjustment scheme for improving the robustness of the approach. Using an automatic matching algorithm, we match resultant images from the RGB-NIR mapping approach and find that its matching result is comparable to the result from matching NIR images, which are always considered as ground truth of vein patterns. In the second approach, we use principles of optics and skin biophysics for uncovering vein patterns. It inverses the process of skin color formation in an image and derives the corresponding biophysical parameters, where veins can be observed. Based on this approach we develop four optical models for simulating skin color formation. They are all based on the radiative transfer equation which quantitatively describes transport of light in the human skin. The first and second optical models use the Kubelka-Munk (K-M) model to approximate the solution of the radiative transfer equation, whose exact analytical solution has not yet been obtained for complex and multiple scattering media such as human skin. In these two models, we assume that the optical properties of human skin is determined by three layers – the stratum corneum, the epidermis, and the dermis, and veins are located in the dermis. To overcome the limits of the first model, the second optical model uses a color optimization scheme and the automatic intensity adjustment scheme. The third and fourth optical models use Reichman’s solution to the radiative transfer equation. In the fourth optical model, we add the fourth layer, the hypodermis consisting of adipose and blood vessels to the skin structure. Because none of the models can provide exact and complete vein patterns, we propose a method to fuse the vein patterns obtained from different models. Experimental evaluations show that the fusion results are much better than any of the single models and also the RGB-NIR mapping approach. Its matching result is even better than matching NIR images. Furthermore, we develop two specific approaches to remove blocking artifacts in JPEG-compressed skin images. The first one is a maximum-a-posteriori (MAP)-based approach which formulates skin image deblocking as an estimation problem, and embeds statistical information of skin images into a MAP model to perform the estimation. The second one is a knowledge-based approach which extracts prior knowledge of skin images from a training set, and uses it to infer original blocks in compressed evidence images. Two inference schemes, a block synthesis algorithm and an indexing mechanism are also proposed for this approach. Both approaches guarantee that the resultant and compressed images have the same quantized DCT coefficients. Experimental results demonstrate that the approaches perform better than other methods. In this research, we break the limit of traditional vein recognition and show its potential for forensic analysis. According to our best knowledge, no one did similar research before.