126 Texture Segmentation Based on Multibands Fusion in Wavelet Domain

In order to overcome the complicated parameter estimation process of model-based texture segmentation methods, a novel multiresolution unsupervised method was proposed in this paper, which implements texture segmentation under the framework of hierarchical clustering in wavelet domain. Firstly, each subband coefficient image is separately clustered to obtain the initial segmentation results using ISODATA algorithm. Then these initial results are fused to obtain the final segmentation result using FCM clustering method with the properly designed inner and inter scale interactions. Experiment on remote sensed image has verified the better performance of our proposed algorithm.