Inhomogeneity test for unsupervised texture segmentation

The authors describe a texture inhomogeneity test method which can be used in model-based unsupervised texture segmentation to increase the accuracy of a model parameter estimation procedure. The method is based on texture properties so that it is able to tackle textures which have the same average tone. It avoids the complicated multivariate hypothesis test by testing a group of uncorrelated texture features individually and combining the results with a combination rule. The texture features used in the test can be extracted easily without any a priori assumption of the texture characteristics. The effectiveness of the test was verified by natural textures