Image-adaptive contrast and entropy based model of regions of visible distortion

This paper presents an image adaptive algorithm for predicting regions of natural images that visually mask distortions created as a result of modification of wavelet domain coefficients. The visual error perception algorithm adaptively predicts the visual detection threshold of distortions added in the wavelet subband. Contrast masking and entropy masking properties of an image are used to generate an activity map indicating the local masking ability of the image. Ground truth data from psychophysical experiments were used to optimize the adaptive activity parameter in order to evaluate the masking thresholds of image regions. The use of this algorithm can increase the amount of watermarking data applied to an image, it is applicable for general wavelet quantization error detection and HVS based compression.