Application Research Of SVM In Image Semantic Learning

With the rapid development of internet,multimedia technologies, image information has been used abroad,an urgent demand has arisen for multimedia information,in which image information plays an important role.Image retrieval generally becomes research hot,traditional retrieval technologies such as text-based image system,aren't able to satisfy demand of people,but content-based imagesystems are based on image visual feature semantic,it can't solve concept-based image retrieval.Retrieval based on image semantic can close to human's image understanding ability.It stands for the orientation of image retrieval.In this thesis,we firstly review the development and the state of the image retrieval research,mainly researching the method of improving image semantic learning effect to extract semantic.Effective features selected using feature selection method in low-level features are tested in SVM(Support Vector Machine) learning.A multiclass classfier creation method of low time,space complexity and good learning effect basing on feature-selection decision tree is found and experimentted in application of SVM. A new extending evaluation method is brought forward in research to learning evaluation criterion.This method considers combination of precision,recall,traditional accuracy evaluation method. As the result, the classifier is better in expressing the ordinary user's concept.Basing on the previous survey, a semantic-based image retrieval experiment system is also presented. This system is an experimental bed for arithmetic testing of features selection and semantic extracting.The thesis describes all kind of functions,query method that system support . Some experiments data are presented.