Quality evaluation of blurred and noisy images through local entropy histograms

Entropy as a measure of information and uncertainty can be calculated in a pixel basis through the use of a spatial/spatial-frequency distribution. A normalized windowed pseudo-Wigner distribution (PWD) and the generalized Renyi entropy have been selected to define locally pixel-wise entropy. The PWD has been calculated in a I-D window basis, adding directionality to the analysis and allowing in such way an anisotropic image evaluation. By means of this, a value of entropy can be assigned to each pixel of the image and therefore a histogram of these entropy values can be obtained. Statistical parameters of such entropy distribution have been derived to define a new image quality metric that can be interpreted as measure associated to the anisotropy of images. The purpose of this paper is to show how such metric constitutes a useful tool to assess both the fidelity and quality of images. Experimental results have been presented for assessing different noisy and blurred images and for the quality evaluation of resolution enhanced images.