Comparative Quality Analysis of Image Global Binarization Procedures

Abstract

This article is devoted to the study of global binarization methods based on the optimization of various quality criteria for the results obtained. Two types of quality criteria are considered, based on the properties of the histogram of the tested image and evaluating the similarity of the latter with the binarized image. The first type includes the criteria used in the well-known OTSU method, the PUN entropy test, as well as the new CHP test. The second type includes the SSIM image similarity test, the W2 gradient test, and the correlated test COR. It is substantiated that for a fixed quality criterion, the binarization procedure is reduced only to searching for a threshold from a certain set of values that maximizes this criterion. It is shown that the criterion PSNR of the closeness of images is unsuitable in this procedure; because the optimal threshold value obtained by this criterion does not depend on the tested image and equal to 127.