On the efficiency of the BRISQUE metric for assessing linearly blurred images when deconvoluted with 3x3 convolution matrices
Artur Giniatullin, Dmitrii Nikolaevich Tumakov, Leonid Alekseevich Elshin · 2024
The issues of correctness of using the BRISQUE metric for assessing image sharpness are addressed. For this purpose, groups of images of three sizes, namely 150x150, 300x300 and 500x500, are selected. These images are deteriorated by horizontal blur and then convoluted with a 3x3 convolution matrix. The matrices are chosen such that their elements surrounding the central element have identical values. Blurred images are reconstructed by such convolution matrices, with elements varying over a wide range. All reconstructed images are evaluated using the BRISQUE metric. In the case of specific images, graphs of the dependences of the BRISQUE values on the values of matrix elements are presented. The extrema of the BRISQUE graphs are shown. It is shown that the BRISQUE can take negative values and its minima may not correspond to sharp images.