Digital image self-recovery using CNN networks

Abstract

This paper proposes a novel CNN-based image tampering detection and self-recovery scheme, this method focus on protecting digital images from forensic manipulation. For this purpose, the watermark algorithm should have two main features first, detecting the tampered zone of the received image by using an authentication watermark; second, recovering the tampered zone by utilizing an image digest watermark. For tamper detection, we propose the CNN-based authentication watermark generation. The image digest is generated using a DCT transform and a jpeg quantization table. “An End-to-End Compression Framework Based on CNN” is used to compress the image digest and improve the quality of recovering the tampered area. We code the compressed image digest by Reed-Solomon error correction code to protect it from a high tampering ratio. Finally, we use Arnold transform for embedding the authentication watermark and image digest watermark in the host image. This embedding method improves security issues. To show the proposed scheme's performance, conduct many experiments. The precision and recall score are calculated to evaluate tampering detection, while for evaluation of image self-recovery, calculate the PSNR and SSIM values of the recovered image. The comparisons with the state-of-the-art methods show that the proposed scheme is superior in imperceptibility, security, and recovery capability.