An objective criterion for the comparison of reconstructed images

The blind deconvolution of images requires the estimation of the unblurred image and the blurring function. Due to its ill-posedness iterative methods are very popular for solving this problem. As a result, several estimated images are available to choose from. In this case, it is important to assess the image quality to ensure a reconstructed image that is visually better. Manual inspection is commonly done however, this is only practical for a few iterations. Objective assessment can also be used but most of these require the unblurred image, which is not available in actual applications. This paper proposes to use the variance and kurtosis of images. This can be accomplished by investigating their relationships with images having various degrees of blur. Experimental results will show that these are consistent for natural images. The proposed criterion is then applied for the selection of the regularization parameters and the final estimated image.