一种改进的稀疏子空间聚类算法

An Improved Sparse Subspace Clustering Algorithm

在现有的稀疏子空间聚类算法基础上提出了一个改进的稀疏子空间聚类算法。首先,利用高维数据可以通过同一子空间的低维数据稀疏地表示这一理论,建立一个稀疏最优化模型,获得稀疏矩阵。然后把稀疏矩阵应用到一个正则化谱聚类算法中,从而有效地把数据聚类到子空间中。最后,该算法应用到一个视频序列中,对每个视频帧里的运动物体进行识别,并与现有的子空间聚类算法相比较。实验结果表明,该算法能够有效地识别运动物体,具有良好的实时性和有效性。

Based on existing clustering algorithms, a modified sparse subspace clustering （SSC） was proposed. Firstly, making use of the theory that high-dimensional data can be sparsely represented by a few points from the same subspace, a sparse optimization model was established and a sparse matrix was ob rained. Then the matrix was applied into a new normalized spectral clustering algorithm. The algorithm can cluster data into subspace efficiently. Finally, this algorithm was applied into a video sequence to identify the move objects. Experimental results showed that the algorithm can identify the move objects with better performance and effectiveness than state-of-the-art algorithms.