Nested Block Based Double Self-Embedding Fragile Image Watermarking With Super-Resolution Recovery

Self-recovery capability is challenging in developing fragile watermarking alongside authentication and tamper localization. In certain situations, authentication and tamper localization alone may not be sufficient. Information about the original condition of the damaged area is essential, particularly in forensic image analysis applications. This study proposes a nested block-based self-embedding method for fragile image watermarking. The data embedded in the cover image includes authentication and watermark bits based on the advanced least significant bit (LSB) method. The watermark bits are generated from the cover images, while the authentication bit is obtained through the SHA-512 operation. Authentication bit embedding is performed using a
$2 \times 2$
block-based Morton pattern. Meanwhile, the watermark bit is embedded twice in different locations within
$4 \times 2$
and
$2 \times 4$
blocks, alternating with the Morton pattern inside the
$4 \times 4$
block using the Zigzag pattern. This approach better preserves the watermark information when tampering occurs over a large area. However, embedding the watermark twice requires more space, necessitating a resize operation before embedding. During the recovery stage, the resized watermark is enhanced using the Feature Super-resolution CNN (FSRCNN) technique. The proposed method demonstrates good imperceptibility quality, with a PSNR value exceeding 40dB and an SSIM exceeding 0.98. Moreover, the proposed method effectively detects and localizes tampering, achieving a True Positive Rate (TPR) of over 95% and a False Positive Rate (FPR) of less than 1% for a watermarked image with 50% tampering. Additionally, the proposed method exhibits outstanding recovery capabilities, resulting in a PSNR of more than 36dB for images with 50% tampering.