Unsupervised feature reduction in image segmentation by local transforms

The automatic segmentation of an image into regions that are homogeneous according to certain properties, such as gray level, texture or color, is an important step in machine vision, and has consequently been the topic of an intensive research . In the general case the segmentation chain includes three steps : Feature extraction, Feature selection, and Segmentation/Classification . The choice for the feature extraction techniques is quite large . Without being complete we refer to [16,21,2] for an extensive exposure of this step . Likewise, several segmentation algorithms have been developed . These can segment images successfully based on multi-dimensional feature spaces in an unsuper-