Kolmogorov-smirnov test for image comparison

Abstract

We apply the Kolmogorov-Smirnov test to test whether two distributions of 256 gray intensities are the same. Thus, this test may be useful to compare unstructured images, such as microscopic images in medicine. Usually, histogram is used to show the distribution of gray level intensities. We argue that cumulative distribution function (gray distribution) may be more informative when comparing several gray images. The Kolmogorov-Smirnov test is illustrated by hystology images from untreated and treated breast cancer tumors. The test is generalized to ensembles of gray images. Limitations of the Kolmogorov-Smirnov test are discussed. © Springer-Verlag Berlin Heidelberg 2004.