Application Research Of Image Retrieval Based On Semantics Of Spatial Semantic Relations And Object Categories

Abstract:In the past of ten year, the search got a maximum evolution, so the image semantic research is more and more seriously in realistic using, to be one of the effective mean to get useful information quickly. Image semantic research now is the hot topic and the difficult point. How to solve the gap between the image characteristics and users of semantic, and to quickly get the users’ ideal image information are the keys. This thesis mainly studies object space relation types and characteristics.Firstly, This thesis introduces and analyzes the background of image retrieval and the current development situation, summarizes the principle, process and CBIR basis technical (namely common color, texture and shape characteristic extraction).Then, in order to clearly understand the process of image retrieval,the research focuses on the color histogram, and develops the image retrieval system by using matlab tools. It prepares for more deeply research of image semantics. Finally, this thesis achieves the image retrieval system based on object categories and spatial relationships, by using threshold segmentation technology,support vector machine technology and relevance feedback techniques. Among the rest, threshold is the typical method of image segmentation methods, it is improved by adding the wavelet transform. And fuzzy C-means clustering which is one of multi-classification methods is improved by adding linear pretreatment of the samples.The experimental results show that color histogram based image retrieval system is simple, effective, reliable. The algorithm improved can get better results and they are more efficient. And this can be a solution to semantic gap between image low-level features and user visual. Also can use in security system to improve its efficiency.