A fuzzy clustering approach to texture segmentation

An unsupervised approach for segmenting texture images is introduced. The following features illustrate the originality of the algorithm. (1) Original observations are converted into uncorrelated local features through Hadamard transformation. The conversion leads to a closed-form expression for the Gibbs distributions associated with the image partition and each texture, and permits the decomposition of the segmentation problem into local statistical decisions. (2) No prior knowledge about the model parameters and the number of textures is required. The unknown information is estimated directly from the observation by means of a hierarchical fuzzy clustering technique. Results are presented for both artificial and real images, showing the good performance of the method.<>