Recently, image clustering methods based on self-supervised learning have achieved great progress, improving both compactness within cluster and separation between clusters. Among them, multi-stage clustering methods introduce pseudo-supervised information to the model through pre-training stage, which brings better performance. But the performance of contrastive learning based clustering methods may be degenerated by the class collision issue caused by false negative pairs. To this end, a novel Prototype representation Learning framework for image Deep Clustering (PLDC) is proposed, which considers the constraints of both cluster and feature training, and reduces the risk of class collision by our designed learning pattern. First, initial feature learning is performed using non-negative sample-pairs, and the higher quality of the features generated, the better performance achieved in subsequent training. Second, a clustering network is trained for generating reliable cluster assignments as pseudo-label information. Third, a prototype-based representation learning refines feature representations and clusters quality. The latter two are iteratively optimized to improve clustering performance. Experimental results on five benchmark datasets demonstrate that PLDC outperforms the state-of-the-art image clustering methods. New levels of clustering accuracies 97.1% and 80.9% are achieved on ImageNet-10 and ImageNet-dogs, respectively.

The performances of semisupervised clustering for unlabeled data are often superior to those of unsupervised learning, which indicates that semantic information attached to clusters can significantly improve feature representation capability. In a graph convolutional network (GCN), each node contains information about itself and its neighbors that is beneficial to common and unique features among samples. Combining these findings, we propose a deep clustering method based on GCN and semantic feature guidance (GFDC) in which a deep convolutional network is used as a feature generator, and a GCN with a softmax layer performs clustering assignment. First, the diversity and amount of input information are enhanced to generate highly useful representations for downstream tasks. Subsequently, the topological graph is constructed to express the spatial relationship of features. For a pair of datasets, feature correspondence constraints are used to regularize clustering loss, and clustering outputs are iteratively optimized. Three external evaluation indicators, i.e., clustering accuracy, normalized mutual information, and the adjusted Rand index, and an internal indicator, i.e., the Davidson-Bouldin index (DBI), are employed to evaluate clustering performances. Experimental results on eight public datasets show that the GFDC algorithm is significantly better than the majority of competitive clustering methods, i.e., its clustering accuracy is 20% higher than the best clustering method on the United States Postal Service dataset. The GFDC algorithm also has the highest accuracy on the smaller Amazon and Caltech datasets. Moreover, DBI indicates the dispersion of cluster distribution and compactness within the cluster.