Optimal Semi-Fragile Watermarking Based on Maximum Entropy Random Walk and Swin Transformer for Tamper Localization

In the multimedia arena, image tampering is an uncontrollable process that necessitates content authentication and tamper detection in a variety of applications. One method that is recommended for meeting all of those needs in the multimedia arena is watermarking. Mobile cameras may now be used to effortlessly take high dynamic range (HDR) photographs, which increase the image’s visual quality and realism. Watermark visibility and recognition algorithms built for standard images may be affected by this introduction of perceptual variations in the image relative to the source. In order to overcome those shortcomings, we introduced a novel, an optimal semi-blind watermarking technique that works for both colour and HDR compressed JPEG images. A unique quaternion dual-tree complex wavelet transform technique is used to extract the highly informative features from the original image. The optimal embedding region in the low frequency sub-band is determined using the maximal entropy random walk (MERW) algorithm. In order to detect tampering and to localize the tampered region a watermark is generated using the swin transformer model and watermark embedding is carried out in the selected optimal blocks. A dual scrambled image is encoded in the effective principal component coefficient values of the singular value decomposition (SVD) Transform in order to authenticate the watermarked image prior to watermark extraction. The semi-blind extraction process is intended to confirm the content’s authenticity by comparing the recovered scrambled watermark with the regenerated original watermark. The process of extraction is merely the opposite of the process of embedding. When compared to previous research, the experimental results demonstrated good imperceptibility with an average PSNR of 65 dB and SSIM of 0.999 and strong robustness against attacks.