Image Source Identification: A Digital Forensic Approach

Abstract

In today’s digital world, with an ever increasing digital crime rate, it is high time that the forensics experts have the tools to map a perpetrator to a crime scene. Source camera identification is a process of associating an image with the source device used to capture it. Camera fingerprint based detection methods as well as machine learning based methods are abundant in the literature to solve the source camera identification problem. However, with the recent advent of counter–forensic techniques, the camera fingerprint based source identification methods have been proven to have vulnerability to counterfeit attacks, which gives the machine learning based methods an edge over them. The machine learning based methods available in the literature follow a basic operating principle: they extract appropriate features from images, train a classifier for camera prediction using the extracted features, and use the trained classifier to predict the source of an image under question. Majority of researches in this domain report a considerably high accuracy as far as prediction is concerned. However, for a forensic expert the model has to be generalized. In the source camera identification problem, the tolerance for false acceptance rate is extremely low, ideally zero. Hence, it is imperative that the model built should predict the source of unknown data with high accuracy. In this scenario, the learning process that a model has undergone, plays the most crucial role, and subsequently affects the accuracy of prediction majorly. In my theses, I discuss various techniques to make an image source identification model learn properly, and establish the importance of concentrating on learning part of a system, through proper interpretation of learning curves. I tested the approaches on the Dresden image database. Our experimental results prove that in this field of research, for fair evaluation and comparison of state–of–the–art techniques, the use of credible benchmark database as Dresden is uncompromisable, as compared to proprietary image datasets.