Detection of Tampered Region Boundaries in Splicing Forgery Images

Abstract

The development of technology caused a significant increase in the use of images in forensic cases. It is common that manipulated images are presented as evidence in courts which requires an authenticity check. In this study, we analyze splicing forgeries where the manipulations are obtained by combining different images. The proposed method divides the original images and the manipulated images into small sub-blocks. After the distinctive statistical information of the images is extracted using ELA (Error Level Analysis), the necessary discrimitative information is learned using a convolutional neural network. The method was tested on the CASIA dataset and is shown to perform comparable or better than some existing methods.