Research On Object And Emotional Semantic Annotation Of Image

Along with the gaining ground of network, the technologies of Memory, Multimedia, Datebase are developing rapidly. The applications'requirements of image are increasing too. There are abundant semantic information in a image. Image semantic annotation is the fundation of all researchs on image semantics. Now there are three ways to annotate the image semantics:artificially tagging, auto-tagging and man-machine interaction tagging. And the last two need the support of a standard material resources base. Thus, researching on the ways of image objects'and emotional semantic annotation and buliding on a image semantic material resources base with high accuracy, have great theoretical significance and applications.Artificially tagging is the basic method of semantic annotation. It has an obvious high accuracy of tagging. But it will be inefficient when tagging mass images because of the massive consumption in manpower and time. The auto-tagging is required to mark rapidly the mass images. Selecting the representative image samples from the mass images to complete the artificially tagging with high accuracy, and building a image semantic material resourse base which has been labeled and described standardly, are important to improve the accuracy of image semantic auto-tagging.We build a experimental model of image semantic layered tagging based on the hierarchical model of image semantics and defined a description frame of image semantic with XML. Structured a description frame on specific iamges' object according to some international classify standard of the object. Then designed a tagging experiment of image object's semantics and emotional semantics with methods in psychological experiment in which inducing an emotion by stimuli. We developed a experiment system to annotate the image object semantics and conducted an experiment system of image emotional semantic with EPrime which is a customized software in psychological study. Finally, we got the data of image object and emotional semantics and analysed the tagging results of emotional model chose in experiment with clustering algorithm to test the validity and veracity. We provided a basis material resourse to special purpose model for researches on image semantics and the association mappings among all levels of image semantic features in the future.We annotated the image semantics with no concern for the influence of personalized factors to understanding a image since limited time. In the future, we can search on the implicit relationships between personalized factors and image emotional semantics, the relevance between visual features and image semantic, and how to reduce or resolve the semantic differences.