The Research On Content-based Image Retrieval

With the popularity of multimedia technology and electronic image equipment,the global digital image information is growing rapidly.The images are scattered around the world due to the lack of effective management,and are difficult to be shared reasonably.Therefore,to manage these images information scientifically and effectively has become an urgent problem.To quickly and accurately search for the needed information in the image database has become a hot research direction among the research scholars,and it is a major challenge for image retrieval.In the past ten years,the image retrieval technology based on content develops rapidly,but traditional text-based image retrieval technology gradually withdraws from the stage of history.The content based image retrieval technology mainly adopts image color,texture,and shape features to jointly promote the research in the database system and computer vision areas.The image retrieval technology is used more and more widely.This thesis focuses on the key technology of content-based image retrieval.Firstly,the research background and significance of CBIR technology,research status and development trend are introduced.With the feature extraction based on low layer,the thesis introduces the relevant algorithms including color feature,texture feature and shape features,and the key technology of content-based image retrieval are discussed in detail.Secondly,the principle and improved method of SIFT algorithm are further studied,and the application of the algorithm in image matching,target recognition and localization are intensive studied.Using the Corel5 image library,ten kinds of image retrieval experiments were carried out on the mentioned two algorithms,and the experimental results were statistically analyzed and concluded.As the precision of the retrieval algorithm to solve the single feature of the traditional high or low speed retrieval problems among multi-feature retrieval methods,the multi feature self-correction weight retrieval algorithm is proposed.With the proposed retrieval algorithm,the retrieval experiment test result shows that the 1000 Corel5 K image database,the algorithm can effectively improve the search,and achieve the desired effect of retrieval.Finally,the whole retrieval experiment is analyzed and summarized,and the improvement points and future research directions are pointed out.