Design and Development of Forensic Techniques for Synthetic Image Identification

Abstract

The huge investment performed by important 3D software development companies in the last few years has taken Computer Graphics Research to a level never achieved before. With the ease of creation and manipulation of digital contents, there is an ever-growing difficulty on differentiating real photographs from computer-generated ones. More than ever, photographs, once a means of secure information, fall under suspicion. Therefore, the identification of computer-generated images has become a difficult, but important, task. Currently, there are several methods in the literature aiming at the differentiation between natural and computer-generated images. Nonetheless, the existing methods lack formalization efforts and validation with real data to effectively bring advances to the forensic community. Toward a more unified and effective classification method, in this project we propose to compare several of the existing methods as well as to design a novel method which can take the main state-of-the-art advances into consideration. The analysis and implementation of current methods will enable us to verify which methods are more suitable in real life cases. We will also consider recent advances in Content-Based Image Retrieval literature to incorporate new image description methods to our proposed solution. In addition, we will perform a study with users in order to evaluate possible new features that we could make use in our classification problem. We will incorporate all the designed methods in our forensics library under development in the Institute of Computing, University of Campinas, which focuses two major potential users: the Brazilian Federal Police and its counterpart in the State of São Paulo.