Unsupervised Texture Segmentation Using Gabor Filters and ICA

Despite the great progress of unsupervised texture segmentation, it still lacks a portable and unified solution to address the diversity of textures. A new framework to gear to the need is presented to take texture features extracted directly by Gabor filters as statistic values. The distinct points are：（1） the framework avoids the difficult problem of selecting parameters of Gabor filters；（2） it integrates texture features using independent component analysis（ICA）；and （3）it treats the independent components as new texture features. Experiments on mosaic natural texture images show the proposed framework gives satisfactory results compared to those based on principal component analysis.