Nonparametric estimation of the number of classes with different average brightness in thermal images

• Published in: Kompʹûternaâ optika Vol. 47; no. 5; pp. 816 - 823
• Main Authors: Galyntich, A.N., Raifeld, M.A.
• Format: Journal Article
• Language: English
• Published: Samara National Research University 01.10.2023
• Subjects: eigenvalues; gram-schmidt orthogonalization; image segmentation; nonparametric algorithm; principal component method; rank histogram
• ISSN: 0134-2452; 2412-6179
• DOI: 10.18287/2412-6179-CO-1284

When there is no information about the number of brightness classes, synthesizing algorithms for automatic image threshold segmentation involves a problem of determining the number of thresholds. The solution to the problem of estimating the number of classes in an image can be based on representing its distribution as a mixture of distributions of brightness classes when priori probabilities are unknown, or estimating the number of histogram modes. At the same time, it is known that the mixture splitting problem has a solution only for certain types of distributions and the histogram modes are not always distinguishable. In the general case, when the distributions of brightness classes are unknown, there are difficulties in applying these methods. The article proposes a non-parametric approach to determining the number of classes that differ in average brightness, based on rank histograms and using the property of local spatial grouping of elements of each brightness class in the image.