Research On Cross-modal Multimedia Retrieval Methods Based On Semantic Matching

With the rapid development of the internet and the popularity of digital electronic devices,people develop from the most single keyword search to image retrieval,and the retrieval of multimedia information such as audio and video.Nowadays people are increasingly demanding information retrieval,and they are not satisfied with the retrieval of single-modal media information.Therefore,the research of cross-modal multimedia retrieval methods is of great significance for the retrieval of internet media information in the multimedia era.The difficulty of cross-modal multimedia retrieval is that the media information in different modalities expressing the same semantics varies in structure.How to effectively and accurately match these heterogeneous data structures is the key to solving this difficulty.To solve this problem,this paper mainly focuses on the problem of semantic matching in multimodal multimedia retrieval.The main research work is as follows:1.For the semantic association of different modal media information,the Ensemble learning method is used to establish a common semantic space based on heterogeneous data.The BaggingSM method is proposed to perform high-level semantic matching for multimedia objects of different modality.Compared with traditional cross-media retrieval techniques,this method has greatly improved the accuracy of search results.2.For the large amount of unlabeled data in the process of training the model is not well used,this article has improved the Bagging-SM method,proposed Semi-supervised learning and Ensemble learning method Semi-RFSM.Firstly,the unlabeled data is predicted and pseudomarked.Then the model is trained using the data with pseudo-tags.This not only solves the problem of a small number of training samples,but also uses the information of unlabeled data.The experimental results show that this method has a higher improvement than the original method in the accuracy of the retrieval results.