[Paper Review] TRLG: Fragile blind quad watermarking for image tamper detection and recovery by providing compact digests with quality optimized using LWT and GA

Behrouz Bolourian Haghighi, Amir Hossein Taherinia+1 more|arXiv (Cornell University)|Mar 7, 2018

Advanced Steganography and Watermarking Techniques22 references3 citations

TL;DR

TRLG proposes a fragile blind quad watermarking scheme using lifting wavelet transform (LWT) and genetic algorithm (GA) for image tamper detection and recovery. It generates four high-quality, compact digests per 2×2 block via LWT and halftoning, optimizes digest quality using GA-based thresholding, and enables four recovery chances per block, achieving PSNR of 46 dB for watermarked images and 24 dB for recovered images after 90% tampering.

ABSTRACT

In this paper, an efficient fragile blind quad watermarking scheme for image tamper detection and recovery based on lifting wavelet transform and genetic algorithm is proposed. TRLG generates four compact digests with super quality based on lifting wavelet transform and halftoning technique by distinguishing the types of image blocks. In other words, for each 2*2 non-overlap blocks, four chances for recovering destroyed blocks are considered. A special parameter estimation technique based on genetic algorithm is performed to improve and optimize the quality of digests and watermarked image. Furthermore, CCS map is used to determine the mapping block for embedding information, encrypting and confusing the embedded information. In order to improve the recovery rate, Mirror-aside and Partner-block are proposed. The experiments that have been conducted to evaluate the performance of TRLG proved the superiority in terms of quality of the watermarked and recovered image, tamper localization and security compared with state-of-the-art methods. The results indicate that the PSNR and SSIM of the watermarked image are about 46 dB and approximately one, respectively. Also, the mean of PSNR and SSIM of several recovered images which has been destroyed about 90% is reached to 24 dB and 0.86, respectively.

Motivation & Objective

• Address the challenge of low recovery quality and limited recovery chances in existing fragile watermarking schemes.
• Improve imperceptibility and security of watermarked images while enabling accurate tamper localization and recovery.
• Optimize digest quality by adapting to block texture using LWT and halftoning, reducing blocky artifacts.
• Enhance security through CCS map-based shuffling and encryption of digests and watermarks.
• Maximize recovery rate under high tampering rates (e.g., 90%) using Mirror-aside and Partner-block techniques.

Proposed method

• Divides the image into non-overlapping 2×2 blocks and classifies each block as rough or smooth using LWT coefficients.
• Applies halftoning to generate two primary and two secondary compact digests per block for four recovery chances.
• Uses genetic algorithm (GA) to optimize threshold values for block classification, minimizing distortion in digest generation.
• Employs a modified Logistic map (CCS map) to shuffle and encrypt digests and watermarks, enhancing security.
• Models watermark embedding as an optimization problem using GA to minimize the difference between original and watermarked pixel values.
• Introduces Mirror-aside and Partner-block strategies to improve recovery accuracy by leveraging spatial redundancy.

Experimental results

Research questions

• RQ1 How can digest quality be optimized in fragile watermarking to improve recovery fidelity under high tampering rates?
• RQ2 Can a quad-watermarking approach with four recovery chances per block significantly enhance recovery success compared to single-chance schemes?
• RQ3 To what extent does integrating LWT and GA improve imperceptibility and security in blind fragile watermarking?
• RQ4 How effective is the proposed CCS map in securing digests and preventing reverse engineering of embedded data?
• RQ5 Can the method accurately detect and recover from complex tampering attacks such as copy-move, collage, and vector quantization?

Key findings

• The PSNR of the watermarked image reaches approximately 46 dB, indicating high imperceptibility and minimal visual distortion.
• The SSIM value of the watermarked image is nearly 1.0, confirming strong structural similarity to the original.
• After 90% tampering, the average PSNR of recovered images is 24 dB, and SSIM is 0.86, demonstrating effective recovery under extreme damage.
• For a specific test case (Girl image), PSNR and SSIM of recovered image after tampering were 32.52 dB and 0.9862, respectively, showing excellent recovery quality.
• The scheme accurately detects and localizes various tampering types, including copy-move, collage, and vector quantization attacks.
• The use of GA for threshold optimization and embedding search significantly improves digest quality and overall system robustness.