Machine Learning-Based Online Source Identification for Image Forensics

Source camera identification is central to multimedia forensics with much research effort addressing to this problem. The assumption made by most existing solutions is that the images are taken from a finite set of camera models available at the classifier training stage. In most cloud-based image services, however, new images are uploaded daily from newly made camera models. To handle these newly made cameras, this paper proposes a new scheme, namely, Online image Source Identification with Unknown camera model (OSIU). In OSIU, firstly, a newly proposed unknown sample triage method detects whether there is a new unknown sample in the image stream. If the images shot by unknown camera models are detected, then a newly developed unknown image discovery algorithm recognizes the samples shot by the unknown camera models. In particular, a new parameter optimization method is invented to maintain good performance of unknown image discovery. Finally, the discovered unknown is incorporated in the identification procedure by employing a customized (K + 1)-class classification, where the additional class consists of the images shot by unknown camera models. Our experiments conducted on the Dresden image collection validate the correctness of OSIU. Regarding the detection of unknown camera models in the online scenario, OSIU generally demonstrates better performance than the state-of-art methods including MSVM, BSVM, CCF, and DBC.