Multi-scale noise estimation for image splicing forgery detection

Abstract

Noise discrepancies in multiple scales are utilized as indicators for image splicing forgery detection in this paper. Specifically, the test image is initially segmented into superpixels of multiple scales. In each individual scale, noise level function, which reflects the relation between noise level and brightness of each segment, is computed. Those segments not constrained by the noise level function are regarded as suspicious regions. In the final step, pixels appears in suspicious regions of each scale, after necessary morphological processing, are marked as spliced region(s). The Optimal Parameter Combination Searching (OPCS) Algorithm is proposed to determine the optimal parameters during the process. Two datasets are created for training the optimal parameters and to evaluate the proposed scheme, respectively. The experimental results show that the proposed scheme is effective, especially for the multi-objects splicing. In addition, the proposed scheme is proven to be superior to the existing state-of-the-art method.