Camera Model Identification for JPEG Images via Tensor Analysis

Camera model identification is to detect by which kind of cameras a photo is captured. A new blind camera model identification method is proposed from the perspective of a composite signal processing, which includes all camera fingerprints, inner embedded algorithms, and external software processing. Take the photo from an actual camera as a tensor, then the residual of Tucker decomposition is related to the nonlinear part of the composite process. Therefore, the pattern implied in the FFT of the decomposition residual is extracted to identify camera model. The SVM classifier is applied to judge whether or not a photo is shot by a certain camera and processed with a series given processing. Photos from five kinds of cameras were used in our experiments to demonstrate this method. Experimental results show that this method has high quite classification accuracy, even when photos were processed after being captured.