Image Clustering Algorithm Based On Regularization Constrained Auto Encoder

Existing image clustering methods rest on a global linear data,which employs prior constraints to estimate underlying subspace of unlabeled data points and clusters them into corresponding groups,thus may fail in handing data with nonlinear structure.Motivated by the huge success achieved by deep learning with its powerful nonlinear representation ability in many applications,in this paper we propose a novel deep image clustering algorithm based on regularization constrained autoencoder.Firstly,the entropy regularization constraint is added to the sparse representation of data,which is used to construct a similarity matrix.the higher information entropy of image is,the more features are extracted from the image.Then the manifold regularization information is combined with the paired constraints,and the joint regularization term is embedded in the autoencoder for training,so as to efficiently use the known supervisory information to guide the clustering process.On the one hand,this method preserves the reconstruction and global features of local manifold structure of the data simultaneously,and on the other hand,the pair constraint rules among known samples are integrated into the target optimization design,which makes the learned low-dimensional features more discriminant and improves the clustering effect of the algorithm to a certain extent.In order to verify the performance of the image clustering algorithm proposed in this paper,the proposed algorithms and others are tested on 4 data sets and analyzed the performance of clustering influence of regularized parameters.The experimental results show that the depth image clustering algorithm proposed in this paper improves the clustering accuracy by about three percentage points compared with the traditional image clustering algorithm,indicating that the features extracted using deep learning are more suitable for clustering;And compared with the clustering accuracy of other depth image clustering algorithms,it is about a percentage improvement,indicating that feature extraction through regularization constrained autoencoder can improve the performance of image clustering.This paper has 22 figures,13 tables and 71 references.