Texture image segmentation using spectral clustering

Abstract

Clustering is a popular and effective method for texture image segmentation. However, most cluster methods often suffer the following problems: need a huge space and a lot of computation when the input data is large. To save the space and computation, we construct a novel algorithm for image segmentation. It consists of two phases: Sampling and clustering. First, we put some detectors into the data space uniformly using orthogonal design method. These detectors can move and merge according to the law of universal gravitation. When the detectors are in a stable status (i.e., do not move), these detectors are used as the representative samples to the next step. Second, to further improve the efficiency and avoid dependence on parameters, the Self-tuning Spectral Clustering (SSC) is used to the representative samples to do the clustering. As a result, the proposed algorithm can quickly and precisely realize the clustering for texture image segmentation.