Research On Related Problems Of Semantic-based Image Retrieval

With the development of computer and multimedia technology, more and more images appear in people’s lives. How to retrieval image quickly and accurately is a research hotspot. Semantic-based image retrieval retrieve the images according to people’s subjective understanding. To a certain extent, it can solve the problem of the subjectivity and difficulty produced in the text-based search methods, and reduce the dependence on underlying characteristics of image in content-based image retrieval. It has become the major tendency in content-based image retrieval.This paper analyzes the key technology of image retrieval at first, describes the image feature extraction, similarity measure and the evaluation criteria of retrieval performance in the method of content-based image retrieval; and introduces semantic model, semantic representation, semantic mapping method in the method of semantic-based image retrieval. The structure of the semantic-based image retrieval system is designed, and the work of extracting underlying feature is completed based on the analysis of the key technology of image retrieval.Then in order to achieve semantic mapping, a method of image semantic classification based on SVM is introduced. Different features or characteristics are used as the input of SVM for different image types. In the stage of sorting the image database, this paper organizes image library by the semantic structure and completes multilevel classification of image database. Then image classification is realized, and all levels of semantic keywords are appropriate for describing the semantic of images.WordNet is introduced in image retrieval to solve the problem of missed or false detection produced by mismatch of the key words in the retrieval process. WordNet is used as a tool to achieve semantic expansion. WordNet is called by jwnl, and the semantic similarity measure methods of Lin and JNC are for calculating the similarity of keywords of the experimental image database. Then right keywords are selected as the expansion of image semantics by the size of similarity to realize semantic expansion.