Quantitative tests-based assessment of biomedical image enhancement procedures

This paper describes a novel method of images enhancement procedures evaluation. A necessity of such method follows from the fact that the results of morphological or statistical image analysis in medical and/or technological applications strongly depend on the effectiveness of image preprocessing. The proposed method is based on standard images called testing sets composed of several basic patterns. Filtered testing sets are compared to basic patterns and the averaged distances between them are used as primary filtering quality scores. Then, they are used to calculation of several secondary parameters called image restoration errors. The image restoration errors make possible separate characterization of filters’ ability to improve image contrast, discrimination of small details or neglect the influence of image parallel shifts on the visibility of image details. Practical application of the proposed method is illustrated by example of comparison of the quality of three exemplary filters: a one based on second-level morphological spectra, Laplace and Sobel, filters. Similar comparison has been performed on the same filters combined with image binary thresholding procedures. At last, the numerical evaluation is compared to visual filters evaluation based on the results of NMR brain image enhancement reached by using different filtering methods.