Splicing forgery localisation using colour illumination inconsistencies

摘要

In the digital imaging era, people used to deliberately distort images or videos for fun or misleading others. Image splicing is one of the methods of manipulation by copying an image from one photograph and pasting it into another. Typically, those two photographs were captured in different environments from various image sources. In this paper, we proposed a simple statistical-based learning-free approach to reveal this type of splicing forgeries using illumination inconsistencies with the assumption that the original images may have uniform illumination. The image first segmented into irregular objects as superpixels and colour illumination is estimated for each superpixel using greyness index in rg-chromaticity space. For each pair of superpixels, the dissimilarity is then estimated. A superpixel region growing algorithm is proposed to extract automatically all the tampered superpixels to localise the spliced region without human involvement. The results of the experiment show that the proposed method effectively localises splicing forgery than the state of art.