Video capturing device identification through block-based PRNU matching

This paper addresses the performance of a PRNU-based (photo response non-uniformity) scheme to identify the capturing device of a video. A common concern is PRNU in each frame being misaligned due to the video stabilization process compensating for unintended camera movements. We first derive the expectation of a similarity measure between two PRNUs: a reference and a test. The statistical analysis of the similarity measure helps us to understand the effect of homogeneous or heterogeneous misalignment of PRNU on the performance of identification for video capturing devices. We notice that dividing a test PRNU into several blocks and then matching each block with a part of the reference PRNU can decrease the negative effect of video stabilization. Hence a block-based matching algorithm for identifying video capturing devices is designed to improve the identification efficiency, especially when only a limited number of test video frames is available. Extensive experimental results prove that the proposed block-based matching algorithm can outperform the prior arts under the same test conditions.