Classification of image pixels based on minimum distance and hypothesis testing

In this article, we introduce a new method of image pixel classification. Our method is a nonparametric classification method which uses combined evidence from the multiple hypothesis testings and minimum distance to carry out the classification. Our work is motivated by the test-based classification introduced by Liao and Akritas (2007). We focus on binary and multiclass classification of image pixels taking into account both equal and unequal prior probability of classes. Experiments show that our method works better in classifying image pixels in comparison with some of the standard classification methods such as linear discriminant analysis, quadratic discriminant analysis, classification tree, the polyclass method, and the Liao and Akritas method. We apply our classifier to perform image segmentation. Experiments show that our test-based segmentation has excellent edge detection and texture preservation property for both gray scale and color images.