Natural image splicing detection based on defocus blur at edges

Abstract

Defocus blur has been used as a cue in image splicing detection. At present, existing methods mainly rely on consistency checking of defocus kernels estimated along suspicious edges (and other reference edges if applicable). However, the texture, nearby edges, light fields as well as noises will influence the information of defocus blur at the natural edges in a certain range, resulting in inconsistent edge defocus blur estimation. As a result, it makes the splicing detection unreliable. In this paper, we analyze the feature of the defocus blur on both the spliced edges and the natural edges, and propose a novel difference-of-defocus-blur based natural image splicing detection method. Compared to the state-of-the-art methods, the proposed method can detect splicing more robustly.