ANANet - MAASNet: A Dual-Stage Framework for Image Tampering Detection and Localization

Abstract

Tampering localization and detection are increasingly indispensable steps with the increasing use of sophisticated manipulation techniques. Herein, we present a two-stage approach that involves Adaptive Noise-Aware Neural Network (ANANet) as preprocessing and Multi-Layer Adaptive Attention Segmentation Network (MAASNet) as segmentation for enhancing tampering detection from images. ANANet effectively removes noise without discarding informative image features so that the regions that have been tampered are not concealed at preprocessing. Following denoising, MAASNet uses multi-layer adaptive attention mechanisms for accurate segmentation of tampered regions and also improves the accuracy of tampering localization. The scheme has the ability to handle most types of image forgeries, i.e., copy-move, splicing, and in painting. The experimental outcome indicates that our scheme is superior to state-of-the-art schemes in terms of detection accuracy, noise variance resistance, and quality of segmentation and hence is an effective solution to image forensics issues.