Study On Content Based Image Retrieval And Application Of Sift Algorithm

Content Based Image Retrieval(CBIR) is a newly emerged technique in the field of multimedia information retrieval.Compared with traditional text based retrieval, CBIR is more descriptive of image content, more objective and automatic and hence gains a very broad application and gradually becomes a hotspot within the multimedia application scope. The Sift algorithm is a new local feature descriptor initially proposed integrally by David Lowe in 2004. Sift feature is highly distinctive and show extraordinary robustness against most disturbances such as scaling, rotation, occlusion, perspective and illumination changes. It's already been proved by experiments that the Sift descriptor holds great outperformance over its peers. The Sift feature is highly useful in scene matching and object recognition, and some begin to apply it to the image retrieval domain.This paper firstly made plenty of survey and study of techniques involved with CBIR and expounded their principles and details. We specially gave implementation solutions on key techniques like feature extraction, multi-feature integration and relevance feedback. Secondly we reaserched the Sift algorithm and made in-depth discussion on its principles and implimentation specifics. Also we studied the application of Sift in image matching and object recognition and proposed a complete set of solutions to all implimentation issues. For the foregoing parts we implimented experimentation systems of CBIR and Sift, and carried out expriments on them. The third focus of this paper is the application of Sift algorithm in image retrieval. We first studied two existing examples and then analyzed the applicability and made a summary. Finally we established a Sift retrieval platform and experimented the retrievability of Sift on different kinds of images. The results were seen to support the forementioned deductions well.