Adaptive keypoint selection for detection of tampering in images and videos

Abstract

Tampering with images and videos for duplicating content and copyright infringement has become a very common problem for original content producers. The main issue with duplication and forgery is that, due to the advancement of forging techniques, it is being increasingly difficult in terms of both computational power and algorithmic complexity to detect and trace the forgeries with good level of accuracy. In this paper, we propose an adaptive keypoint based approach to detect the presence of forgery in images. Our approach is independent of the input dataset, and provides good level of accuracy for forgery detection. The system is tested on REWIND dataset, and an accuracy of more than 85% was observed. Our approach can be further extended to incorporate machine learning in order to improve the accuracy.