Restoring blood vessel patterns from JPEG compressed skin images for forensic analysis

The recent development of forensic methods makes criminal and victim identification based on blood vessel patterns hidden in color images possible. The performance of these methods highly depends on the quality of input images. The JPEG method, being the most popular image compression method, generates blocking artifacts, which seriously degrade image quality and make uncovering blood vessels difficult. This paper proposes an algorithm to restore blood vessel patterns hidden in JPEG compressed skin images. An analysis is first performed to identify critical discrete cosine transform (DCT) coefficients that control the quality of blood vessel patterns. According to the analysis, an algorithm is designed to restore these coefficients by exploiting DCT coefficients in different blocks and channels of JPEG compressed images. A blood vessel pattern matching method and a database with 978 images are employed to evaluate the effectiveness of the proposed algorithm. The experimental results demonstrate that the proposed algorithm can effectively alleviate blocking artifacts and restore blood vessel patterns for matching and that it outperforms the knowledge-based deblocking method, which is specially designed to restore skin images.

I. Introduction

Many biometric traits, such as face, fingerprint, DNA, and palmprint, have been widely used in legal cases for criminal and victim identification. They are not applicable to evidence images with only non-facial skin. These legal cases include but not limited to child pornography, violent protests, and terrorist attacks, where criminals always hide their faces. For example, the convenience of the Internet has promoted a mass increase in child pornography. The U.S. Department of Justice reported that there were more than 20 million IP addresses nationwide offering child pornographies and the demand for child pornographies was responsible for the increase in child sex abuse cases [1]. It is believed that most of these illegal images are likely taken with consumer cameras and compressed with the JPEG method, which is widely installed in them. Though tattoos and skin marks have been considered, they have some weaknesses. Tattoos are neither universal nor unique and distinctive skin mark patterns are not always available. Blood vessels under human skin form large and robust patterns providing rich information for personal identification. They are generally regarded unique and stable over time, except for the cases of injury, disease (e.g. varicose veins and spider veins), and surgery. Thus, using blood vessel patterns for criminal and victim identification is desirable. Based on the principles of optics and skin biophysics, Tang and Zhang et al. [2], [3] proposed methods to uncover blood vessel patterns hidden in color skin images for criminal and victim identification. These methods were evaluated on uncompressed images and the experimental results were encouraging. With these methods, blood vessel patterns can be extracted from color skin images for personal identification. However, the quality of blood vessel patterns generated by them highly depends on the quality of input images. The JPEG method can severely degrade image quality and make uncovering blood vessel patterns difficult. Figs. 1(b) and (d) show two blood vessel patterns, one from the uncompressed image given in Fig. 1(a) and the other from its compressed version with a quality factor of 50 given in Fig. 1(c). Fig. 1 indicates that the resultant blood vessel patterns from [3] can be highly influenced by the JPEG method. Fig. 1 Uncompressed image, JPEG compressed image and the corresponding resultant blood vessel patterns from [3]. (a) is an uncompressed image and (b) is the blood vessel patterns from (a), (c) is the JPEG compressed version of (a), and (d) is the blood vessel patterns from (c).