Machine Learning Based Saliency Algorithm For Image Forgery Classification And Localization

Abstract

In this paper, two algorithms are proposed to classify and localize image forgery. In the first algorithm, deep learning based convolution neural network is used to classify spliced and authentic images. In the second algorithm, machine learning based saliency algorithm is used to detect and localize forged images. Input images are preprocessed using color illumination maps with equal size and channels. Saliency algorithm detect unique features such as color illumination, pixel resolution etc. of the image. These unique features depict the forged regions in an image. The results are obtained on CASIA-v1, CASIA-v2, DVMM and BSDS-300 dataset. Simulated results in both algorithm are better as compare to the state of the art method.