Abstract: A novel local saliency features of interest extraction algorithm based on visual saliency measurement theory is provided.Firstly,the focuses of attention are extracted by an iteration process,which is fitted to the attention shift mechanism.Then,we treat it as local saliency points,and the saliency edge points are got according to the combination of Canny edge graph and saliency map.After that we employ main hue,texture features of saliency points and edge orientation histogram features of edge points as feature vector for interesting objects representation.This algorithm emphasizes the salience areas in the scene,so that it can improve process efficiency and fit to human visual perceptual characteristic,otherwise this multi-channel feature describe method ensure the algorithm’s adaptive to various scenes.