Query Translation Based On Visual Information For Cross Language Retrieval

With rapid development of the Internet,the multilingual information on the web has increased dramatically.In that case,retrieval results of their own language are always difficult to meet the requirement of users.As one of supporting technology in cross language information retrieval,query translation has important research value and application prospect.Because traditional query translation is mainly based on text information with ambiguity and hard to achieve query disambiguation during translation process,this paper proposes and implements a new query translation method based on visual information as follows:1)Data set building.The data used in this study is a set of query words.The method of query translation based on visual information is to convert the query words of the text information into visual information.Through the image retrieval system,we can get the image set of each query word and complete the transformation of information and the initial construction of the data set.2)Query translation.This paper proposes a method of using image visual information to replace text information during query process.According to existing researches,when the image sets of a candidate translation and query are similar to each other,they may refer to the same thing.The relevance between the query word and the Internet data,which is obtained from the image search,is so different that some of the query word correlation is obviously higher than the others.,which are considered as noise and need further processing to become more accurate.With the method of similarity measurement,the result of query translation can be obtained from the image set which is closest to the query word.3)Experimental verification.The research method is verified by the experiment.The experimental results show that the accuracy of query translation is 60.5% with only visual information.Besides,our result has obvious promotion compared to the dictionary translation and existing translation system,e.g.,Baidu and Google translation.With respect to traditional method,query translation based on visual information doesn't need context of query and additional text information.Above all,this method has the value of further research to get more accurate results.