Low Quality Image Forensics Based on Imaging Constraints
Abstract：In this project, we try to improve the robustness and accuracy of image tampering forensics method. We mainly focus on forensics of low quality images. We propose to explore the constrains of scene and imaging condition, because the traditional statistical features are almost failed for these images. Based on the auxiliary information or priori knowledge, some 3D information and camera parameters can be restored. Thus, the constraints, such as lighting or shadow directions consistency, geometric constraints of some known points, and object size, should be expressed more accurately with the help of the restored information. Though 3D reconstruction from a single image is an ill-defined problem, by regulating with 3D object data collected by Kinect, we can reconstruct the 3D information from the image with a morphable model. To make the image forensics more practical in terms of robustness and accuracy, we try to solve the contradiction and compatibility problems of multi forensics clues by a decision fusion method. Besides, we also consider the problem caused by the measurement precision of forensics clue.
Keywords: image forensics; constraint of imaging environment; decision fusion; auxiliary information; Kinect
Contact:
WANG Wei
E-mail: wwang@nlpr.ia.ac.cn