Adaptive Multimodal Robust Feature Learning Based on Dual Graph-regularization

In the era of big data, the widespread existence of massive multimodal data has brought about great changes in the characteristics of data: there are many types of data and low value density. Different types of data both play an independent role and complement each other. Discovering the hidden value behind multimodal data has become a The key to big data mining. Aiming at the low quality of multimodal data, this paper proposes a new multimodal robust feature learning method. This method effectively reduces the influence of noise data on fusion results by introducing modal error matrix. In addition, the double-graph regularization mechanism of data manifold and feature manifold is designed to describe the double spatial structure of modal data to ensure the stability of data in the fusion process. In six practical On the multimodal data set, based on the three evaluation indicators of accuracy (Accuracy, ACC), normalized mutual information (Normalized Mutual Information, NMI) and purity (Purity, PUR), it is c