Unsupervised segmentation of texture images using feature selection

Highly precise segmentation of texture images is a technique which is indispensable in image understanding and image recognition. This paper proposes a method which represents the features of the texture image by using the wavelet transform, the nonlinear transformation, and a Gaussian filter, which are used in time-frequency analysis. A highly precise segmentation procedure for texture images is presented in which a feature suited to segmentation is selected from among a number of features. The usefulness of the proposed method is demonstrated through experiments using artificial data, and a result for a natural image is presented as an application to a real problem.