Clustering of Image Covariance Matrixes on Lie Group Manifold

An image clustering method based on covariance matrix and mean-shift algorithm on Lie group manifold is proposed. Firstly, according to the extracted multidimensional correlation features of image, the covariance matrixes are calculated to form a Lie group manifold. Secondly, by using the mapping relationship between Lie group and Lie algebra, the steps of covariance matrixes clustering based on mean-shift algorithm on Lie group are established. Finally, the example verifies that the mean-shift algorithm on Lie group manifold can better obtain the clustering information of the image covariance matrixes, and the average accuracy of image classification is 95.80%, which improves accuracy by 3.1% compared to traditional algorithms. Moreover, if the kernel function, bandwidth and threshold are set up reasonably, the image clustering is more efficient and accurate. For the unit Gaussian kernel function, the optimal bandwidth is 1.2, the optimal threshold is 0.8. It provides an algorithmic basis for the application of Lie group machine learning in high-precision automatic target clustering.