| Content-Based Image Retrieval(CBIR,Content Based Image Retrieval) is the area of multimedia information retrieval,a new technology Operation.It is because of the traditional text retrieval in(TBIR,Text Based Image Retrieval) existence some defects:(1) Characterization of the image detail is not sufficient;(2) Subjective and non-Uncertainty;(3) Marked the completion of the process.Because of these defects,CBIR techniques have come into being.Compared to the TBIR technology,CBIR has so many advantages:(1) Directly detect information from the image content,Automatic analysis by computer itself,no manual participate,thus avoiding the subjectivity of manual tagging and greatly Increased efficiency and automation;(2) Detection is an approximate matching rather than exact match.The purpose of CBIR is to identify with the query image(query image) the most similar images,the system returns results based on a list of the degree of similarity,which is obviously different with traditional method.(3) It is an interactive search,relevance feedback process,the user can provide feedback to the seizure Continuously improve the way claims and amendments to improve the retrieval accuracy.CBIR retrieval technology can be seen Full description of the image content,objectivity,and automatically and efficiently Etc.CBIR has a very broad space of application.Concerned widespread,CBIR is becoming a hot spots in the field of multimedia applications and technology research.In the image retrieval Image Registration is one of the most basic and most critical issue,which originated from a number of practical problems in many fields such as access to information of different sensors Integration;different times under different conditions was the difference between the image side of the supervisor; imaging systems and objects scene changes in the images under the three-dimensional information acquisition;image in target identification;multi-media physical databases and so on. SIFT algorithm,completed by David Lowe in 2004,is a new local feature descriptor. SIFT descriptor has many characteristics:(1) Distinctiveness,informative for the characteristics of the massive database into the Quick and accurate matching.(2) SIFT feature is the image of the local characteristics of translation,rotation,zoom scale,the brightness changes over Block and has a good noise invariance,the perspective changes,affine transformations have also maintained a certain degree Stability.(3)Massive,even a small number of several objects can also have a large number of SIFT feature vectors.(4) relatively fast speed,the SIFT matching algorithm optimization can be achieved even in real-time requirements.(5) scalability can be very convenient with other forms of joint eigenvectors.Mikolajczyk done in partial characterization of a series of sub-performance comparison test,SIFT Operator has been proven to match the performance of its significantly higher than the same type of local special Levy.Sift in the scene matching algorithm,target recognition in areas such as the use of a great value,its Applications in the field of image retrieval research has already begun.This article first CBIR system based on the points involved in the characteristics of wide-baseline image matching technology to Presentation and in-depth study of the SIFT algorithm and implementation details in the image of their match,Target recognition and positioning applications in areas such as a serious discussion.And for the SIFT algorithm for Large-capacity database image retrieval when the efficiency of the bottleneck,given previous study Principal component analysis of descriptors,such as dimensionality reduction method,this paper DBSCAN clustering algorithm In this paper,a solution,that is,clustering SIFT feature points to match the type of class to replace the original point-to-point matching algorithm to reduce the number of matches in order to achieve search efficiency. Search experiment at the same time,a very good experimental results supported the above conclusions. |
PDF Full Text Request