Incremental Clustering With Hard Centers

FU-PCM is an effective clustering mode by inducing the regularization constraint in C-means to avoid the interference of noises and outliers. However, it is difficult to obtain the satisfactory performance in the context of imbalance category distribution. To address above problem, first, we propose a novel clustering model, which exploits Pearson Correlation Coefficient to auto-balance the optimal equation according to the category distribution. Then, we extend it to the incremental model, which only learns samples at the current frame to update the original model by mapping centers of two adjacent frames to the distinguishable space, and mining hard centers to recognize new categories in online datasets. In this article, offline and online methods are verified in popular datasets, and experimental results demonstrate the effectiveness and efficiency of the proposed models.