Research On The Key Technology Of CBIR Based On Feedback And Multi-feature

With the rapid development of image acquisition technology, storage technology, and computer network technology, the image data showing rapid growth in geometric form, and there have been many high-volume image database on the Internet. Thereby also to the image management and retrieval brought unprecedented difficulties. Especially content-based image retrieval (CBIR) based on the looming image retrieval technology. CBIR refers to the image database retrieval based on the underlying features of the image, retrieve the query image matching images.Image Retrieval field is currently a large problem:There is a big difference that is the "semantic gap" between the extracted single feature can not accurately describe the image information, and the underlying image features and image content for people to understand. Feature fusion can solve a single feature can not fully describe the problem of image information; Effective way to solve the semantic gap is related to the introduction of a feedback mechanism,In this paper, the color, texture, shape and other characteristics of combined relevance feedback technology to image retrieval, Designed and implemented a complete a CBIR system. The main contents are as follows:(1) Firstly, starting from the texture feature, using GLCM and homogenization LBP texture feature extraction algorithmr, respectively, the Euclidean distance and histogram intersection method similarity matching, design and implement image retrieval system based on texture features MATLAB7.1 internet.(2) We can not fully express the image information for a single feature of this shortcoming, and the use of a custom weighting method features the image of the underlying texture, color, shape, fusion, and implements a feature fusion based image retrieval system.(3) To introduce relevance feedback mechanism based on support vector machine in the CBIR system, human-computer interaction is implemented, and improve the retrieval satisfaction.In this paper, we tested the CBIR system on the international standard image database Corel, and obtain good results.