一种面向视觉注意区域检测的运动分割方法

摘要

In order to detect the visual attention region more effectively,this paper proposed a novel motion segmentation method using visual saliency detection.The method clustered the feature point trajectories using a two-stage clustering algo-rithm.It pre-clustered the most stable trajectories using Mean-Shift in the first stage.Then,the paper proposed an unsuper-vised clustering method to cluster the trajectories and detect the number of motions automatic.At last,it generated the motion saliency map with the segmented spare feature points.Experimental results show that the proposed method is capable of achie-ving both good accurate and the stable performance.%为了更有效地检测出视觉注意区域，结合视觉显著区检测的特点，提出一种面向视觉注意区域检测的运动分割方法，该方法用一种层次聚类方法将特征点的运动轨迹进行聚类。首先用均值偏移算法扩大了不同类型运动之间特征向量的差距，同时缩小了相同运动类型的差别；继而，用一种无监督聚类算法将不同类型的运动进行分割，同时自动获得运动分类数；最后利用运动分割结果，提出一种结合空间和颜色采样的运动显著区域生成方法。与以往方法相比，该方法能够将不同类型的运动自动进行分割，生成的视觉注意区域更为准确，而且稳定性大幅提高。实验结果证明了该方法的有效性和稳定性。