Nonparametric Method of Estimating Number of Classes in Image Segmentation
Roman V. Podrezov, M. A. Raifeld · Optoelectronics Instrumentation and Data Processing · 2020
Abstract One of the important problems of automatic threshold image segmentation by brightness are questions of the number of brightness classes and, as a consequence, the required number of thresholds. The solution to the problem of estimating the number of classes in an image is often based on representing its distribution as a mixture of distributions of brightness classes. It is known that this problem (splitting a mixture) has a solution only for some types of distributions, and is difficult to apply when the distributions of brightness classes are unknown. This paper presents a nonparametric method for determining the number of classes based on rank histograms and using the property of the local spatial grouping of elements of each brightness class in the image. Comparison of the proposed method to different criteria for assessing the number of classes in images showed it to be effective.