TRLF: An Effective Semi-fragile Watermarking Method for Tamper Detection and Recovery based on LWT and FNN

This paper proposes a novel method for tamper detection and recovery using semi-fragile data hiding, based on Lifting Wavelet Transform (LWT) and Feed-Forward Neural Network (FNN). In TRLF, first, the host image is decomposed up to one level using LWT, and the Discrete Cosine Transform (DCT) is applied to each 2×2 blocks of diagonal details. Next, a random binary sequence is embedded in each block as the watermark by correlating DC coefficients. In authentication stage, first, the watermarked image geometry is reconstructed by using Speeded Up Robust Features (SURF) algorithm and extract watermark bits by using FNN. Afterward, logical exclusive-or operation between original and extracted watermark is applied to detect tampered region. Eventually, in the recovery stage, tampered regions are recovered by image digest which is generated by inverse halftoning technique. The performance and efficiency of TRLF and its robustness against various geometric, nongeometric and hybrid attacks are reported. From the experimental results, it can be seen that TRLF is superior in terms of robustness and quality of the digest and watermarked image respectively, compared to the-state-of-the-art fragile and semifragile watermarking methods. In addition, imperceptibility has been improved by using different correlation steps as the gain factor for flat (smooth) and texture (rough) blocks.