Research On Multi-media Database Retrieval Basing On The Chinese Natural Language

The informationization of education has brought forward new requirements forteachers. As the supervisor of the learning resources, teachers are to fix the mediaresources that are needed, and provide the query clues for students. Although there areabundant multi-media resources in the Internet, the existent searching retrievals aremainly based on the keywords, which searching accuracy is not so high. Teachers ofmiddle and primary schools are busy doing their routine jobs, and they are various inapplying the technology of computer network. Therefore, they need a convenient andfast system to search the multimedia resources.Natural language query means that the users describe the query object in theretrieval system, which extracts the query requirement and the key features. Then theretrieval system feedback the query results according to fixed rules and algorithm.The Chinese natural language based multimedia retrieval mainly include threeprocedures: extracting the key features of the object media from its Chinese querywords;searching records in the multimedia database which meet the queryrequirement and have the higher conformity;providing the records to the usersaccording to its conformity.In this thesis, we research into the features of the Chinese language understandingand general ways of word division, found a word division system for our own use,divide the query texts and label its parts of speech. After omitting the function wordsand the default words from the query texts, we can get the description of the objectmedia and call them "theme content". Besides, we extract the color words from thetheme content basing on the color dictionary, and combine those that the users inputto be the main tonality words;extract the object words and its attribute words basingon the main body dictionary and main body attribute dictionary;and extract thebackground words only if there are structures such as "The background is".Furthermore, we should extend tonality words according to the synonym dictionarybefore calculating the conformity.We adopt the conformity to judge the distance between the object media andmedia in the database. The media include text features and content features, but wemainly refer to the content while calculating the conformity. We found expressionmodels of content feature for the image, flash, video and audio, and have differentways to calculate conformity for different content features. We find the number ofsame words in the extended tonality words and content description records tocalculate the conformity. We change the color words to HIS model, and calculate theconformity of the tonality which is labeled by numerical value. We calculate theconformity by comparing the main body records with the object and its features. Afterall the content features are fixed by it importance, we calculate the total conformity,and feedback the most similar twenty query results to users.Basing on the above, we design a Chinese natural language based multi-mediaretrieval system. After registration, users can input the Chinese natural query texts andchoose the file format, size and media type. After extracting the tonality content andthe content feature, and calculating the conformity, the system will feedback theresults to the users according to the magnitude of conformity, which include the filesize, the conformity and its website. The experimental results suggest that the systemis simple, can divide words, can extract the tonality, contents and feature words,which is just through a few query texts. The accuracy is very high for the recordswhich are precise in the multimedia database, proving the content feature basedretrieval. At the end of the thesis, we conclude all our research and bring forward thefuture research orientation.