Deep image clustering using ensemble learning and multiple features of deep neural networks

Deep learning is a powerful tool in clustering complex and large images.Most of the deep image clustering methods perform clustering based on the representation vectors obtained from a deep neural network training, so that the features extracted from the last layer of the network are used for the final clustering. Benefiting from different semantic information extracted from several deep networks can play an effective role in improving the efficiency of the clustering results. In this paper, we present an approach based on ensemble deep clustering, where by linking deep clustering methods and ensemble learning, we use multiple deep neural networks advantages together.In this regard, five autoencoders with several convolutional layers are trained, in each of which the transfer learning is performed to improve its accuracy and performance. After learning the effective representations of the images by different deep models, these vectors are clustered and their results are combined according to the ensemble approach. Then,the final clustering is calculated using the common information of the base clusters. The results of the proposed method on four standard image datasets represent its more effective performance rather than the recent deep image clustering methods.