Correlation Recognition Between Image And Text In Weibo Based On Multi-Source Features

With the rapid development of society and economy,reading images era has come.The application with image as the research object is becoming more and more popular.However,these researches have a common problem,"semantic gap" between image low-level features and high-level semantic features,and this problem is always a difficult point in image analysis and understanding.With the appearance of social media,Weibo has been paid much attention to,and the data of Weibo show in the form of not only text but also text around multi-source information,such as images and social information.The purpose of this paper is to focus on the correlation between the image and text using the multi-source information in Weibo microblogging.For Weibo data,the features of multiple source information,image features,text features and social features,are extracted in this paper firstly.Secondly,according to these features,three methods,the identification of correlation between image and text approach based on BP neural network in Weibo,the identification of correlation between image and text approach based on SVM in Weibo and the identification of correlation between image and text approach based on feature mapping in Weibo,are presented.The BP neural network and the SVM training models have been used in these approaches,besides,the method based on feature mapping using the genetic algorithm to eliminate the heterogeneity between image features,text features and social features.In this paper,accuracy of each kinds and the general accuracy are used to evaluate experimental results.The experimental results show that the method based on feature mapping is more effective than the rest two methods in related prediction and whole prediction,and the method based on SVM method is more effective than the rest two methods in unrelated prediction.