A study of the effect of image quality on texture energy measures

Abstract

Assessment of the quality of a digitized image is examined using an analytic model of the modulation transfer function of the image processing system. It is demonstrated that there is a significant and quantifiable effect on texture energy measures, as defined by the particular cases of the Laws convolution masks and a new adaptive approach to spatial filtering. In particular, the radius of the best-fit Gaussian to the point-spread function correlates inversely with the power of local operators to discriminate visually similar textures. Results indicate that the use of adaptive convolution masks improves discrimination significantly, and is more tolerant of instrumental variations.