Salient texture contrast difference based blur metric

This paper presents a no-reference salient texture contrast difference based blur metric with integrated simple visual attention weights for a better simulation of the HVS response to blur distortion. The proposed blur metric is modeled as an exponential function of the salient texture contrast difference which is defined as a weighted sum of the salient texture contrast difference of each patch between the two blurred versions of the input image. The experimental results on public available databases show that the proposed method is monotonic, robust to additive noises, and also consistent with human visual system.