矩阵的低秩稀疏表达在视频目标分割中的研究

发布时间：2018-04-14 23:11

本文选题：能量最小 + 图分割　；参考：《电子科技大学学报》2017年02期

【摘要】：提出了一种视频目标跟踪与分割的在线算法。该算法将每帧图像中的超级像素作为一个数据点,根据已知的目标和背景建立模板,当前帧中待分割的目标可以看成已知模板的稀疏线性表达。根据此线性表达的系数可以建立描述当前帧与模板的相似性矩阵,即表达子。由于视频图像的连续性,表达子具有低秩和稀疏的特征。因此通过求解矩阵的低秩稀疏的优化问题可以得到当前帧中所有数据点属于目标的概率分布。为了获得基于像素级的分割结果,通过能量最小框架,并利用图分割的方法最终实现目标的分割。实验结果表明该算法具有良好的分割效果。
[Abstract]:An online algorithm for video target tracking and segmentation is proposed.The algorithm takes the super-pixel in each frame as a data point and establishes the template according to the known target and background. The target to be segmented in the current frame can be regarded as the sparse linear representation of the known template.According to the coefficients of the linear expression, the similarity matrix describing the current frame and the template can be established, that is, the expressor.Because of the continuity of the video image, the expressor has the characteristics of low rank and sparse.Therefore, the probability distribution of all data points in the current frame can be obtained by solving the low rank sparse optimization problem of the matrix.In order to obtain the segmentation results based on pixel level, the goal segmentation is achieved by using the energy minimum frame and the graph segmentation method.Experimental results show that the algorithm has a good segmentation effect.
【作者单位】：电子科技大学通信与信息工程学院;成都航空职业技术学院航空工程学院;电子科技大学电子工程学院;
【基金】：国家自然科学基金(61271288) 教育部博士点基金(20130185130001)
【分类号】：TP391.41

【相似文献】