Tamper Detection in Watermarked Images Using NROI-Based Circular-Harmonic-Fourier Moments

Abstract

In view of the lower discretization of image tampering watermark, the overall block-level operation characteristics are easy to trigger the square effect. The watermark-containing region has the attribute of multiple sub-blocks, there is a watermark-containing region with poor perceptibility, the watermarking information is not visible. The interference resistance and robustness are poor, for this reason, we propose a tamper detection method in watermarked images using NROI-based circular-harmonic-fourier moments. The NROI sub-block is selected, the circular-harmonic-Fourier moments of each sub-block are calculated. The smoothing sub-block in the polar coordinate system is decomposed orthogonally, the geometric invariance of the circular-harmonic-Fourier moments is embodied in the magnitude of the moments. The low-order circular-harmonic-Fourier moments are selected to superimpose and sum up the products of the circular-harmonic-Fourier functions of the same level, so that the reconstruction of image is realized; Take the sub-block circular-harmonic-Fourier low-order moments of the image reconstruction result in the NROI region as feature vectors. Determine the robust transformation coefficients by defining the robust coefficient selection rules to obtain the watermark tampering location to avoid the occurrence of the square effect; The complete watermark sequence is extracted by comparing the amplitude variation of all the moments with the domain values. The extracted binary watermark sequence is compared with the original watermark stored in the NORI location to verify whether the image has been tampered or not, through the implementation of the corrosion operation and expansion operation, the image region that matches the actual tampering is sensed. The experimental results show that this method can resist attacks such as white noise, rotation, and scaling for image samples with strong concealment of watermark images, and maintain clear and distinguishable watermark information. The PSNR of watermark information is higher than 40, SSIM index is higher than 0.9, and JND index is lower than 5, demonstrating excellent anti-interference and robustness. The detection accuracy of this method for tampering with image watermark information under different attack states is higher than 94%, the detection time is less than 8ms, and the detection efficiency is relatively good.