Adaptive image restoration using a perception based error measurement

This paper deals with image restoration; we have developed a novel, perceptually inspired image restoration method which takes human perception knowledge into consideration to reverse the effects of blur. Instead of using a conventional greyscale based error measurement such as the MSE, we compare local statistical information about regions in two images using a new error measure. The new method provides a better appraisal of image quality in terms of human vision. We extended the popular constrained least square error cost function by incorporating this novel image error measure. Using the well known Karush-Kuhn-Tucker theorem, we have mathematically verified that there exists an optimal solution to this nonlinear constrained optimization problem in terms of the Hopfield neural network. We show that the new restoration algorithm visually restores images as well as the previously presented LVMSE-based algorithm.