DYNAMIC RESIDUAL FORGERY-AWARE DENOISING AND GRADIENT-AWARE PATCH SALIENCY FOR ROBUST IMAGE FORGERY DETECTION

Abstract

There has been an increase in the use of digital image tampering due to the availability of advanced image editing software, which has become a significant challenge for the authentication of images and digital forensics. Preprocessing and effective segmentation procedures are required to detect the tampered regions as well as store useful forensic information. The research proposes robust image forgery detection architecture in the integration of the Dynamic Residual Forgery-aware Noise Suppression (DRFNS) for denoising and Gradient-Aware Patch Saliency (GAPS) for segmentation. The DRFNS algorithm works adaptively patch-based denoising by calculating residual signals and other indicators of forgery like edge density and structural inconsistencies. The method allows high noise suppression in normal areas and gives hints of manipulation, which are subtle in suspicious areas. Gradient variation, texture anomaly and colour anomaly are then examined and used to compute the patch-level saliency scores on the denoised image by the GAPS segmentation technique. Additionally, K-Means clustering is also included to improve the accuracy of segmentation by having the ability of dividing suspicious regions and normal areas effectively. As shown by experiments, the proposed framework has obtained the highest denoising results by achieving a PSNR of 40.48, SSIM of 0.96, and RMSE of 0.0093. In the case of segmentation, the approach has an accuracy of 97.00% and it is outperformed the existing approaches. These findings prove that efficacy of the proposed framework to trustworthy digital image forgery identification.