Cross-scale feature fusion-based JND estimation for robust image in DWT domain

Owing to the rapid development of just-noticeable difference (JND) and the high robustness of discrete wavelet transform (DWT) based watermarking, robust watermarking methods based on JND models have been extensively adopted in the area of image copyright protection. In general, in existing JND model-based methods, transform domain-based features play an important role during the quantization strength measurement. However, most existing methods only use the subband-specific features from the final multi-scale wavelet transform domain, ignoring the feature relevance among subband-specific features with different scales among RGB channels. To address this issue, in this study, we put forward a JND estimation based on cross-scale feature fusion (CFJnd) mechanism in quaternionic DWT domain for robust image watermarking. For the proposed CFJnd, we first design different quaternionic wavelet scales for the JND modeling, which can be used to measure the minimum distortion level of the embedding-specific subband relevance. Furthermore, the cross-scale features among different subbands are adopted to jointly generate the pattern, texture and color measurement, making the final perceptual masking more representative. In addition, we further develop a CFJnd-based robust image watermarking, and the watermark is embedded by quantifying the candidate quaternionic DWT coefficients of the embedding-specific subband on a level which is adaptively defined according to the proposed CFJnd. Experimental results based on the publicly available image database show that the proposed method can produce reliable and promising results, compared to other state-of-the-art image watermarking methods.