Skip to Main Content
SPN (sensor pattern noise) has been proved as an effective approach for source cameras forensics. But, due to the limitation of existing methods, the SPN extracted in images are heavily contaminated by the scene details, and misidentification rate is high unless large size images are tested. In this paper we propose a novel approach for the extraction of SPN to improve the identification accuracy. Firstly, the images are analyzed by orthogonal wavelet transform, then used with edge-preserving bilateral filtering for approximation sub-band, and adaptive minimum mean squared error filtering for detail sub-band respectively, de-noising by bilateral filtering in spatial domain. In this way, SPN in different frequency components can be extracted effectively. Afterwards, the identification for extracted SPN has been achieved using classifier based on maximum correlation principle, the impact on identification are discussed under different JPEG factors.