An improved sensor pattern noise estimation method based on edge guided weighted averaging

Abstract

Sensor Pattern Noise (SPN) has proven to be an effective fingerprint for source camera identification. However, its estimation accuracy is still greatly affected by image contents. In this work, considering the confidence difference in varying image regions, an image edge guided weighted averaging scheme for robust SPN estimation is proposed. Firstly, the edge and non-edge regions are estimated by a Laplacian operator-based detector, based on which different weights are assigned to. Then, the improved SPN estimation is obtained by weighted averaging of image residuals. Finally, an edge guided weighted normalized cross-correlation measurement is proposed as similarity metric in source camera identification (SCI) applications. The effectiveness of the proposed method is verified by SCI experiments conducted on six models from the Dresden data set. Comparison results on different denoising algorithms and varying patch sizes reveal that performance improvement is more prominent for small image patches, which is demanding in real forensic applications.