| With the rapid development of multimedia technology, computercommunications technology and the Internet, image has been favored by people.However, currently, a large number of images is still showing the distribution stateof disorder so that the useful information contained in the image cannot beeffectively utilized. It requires a technology to locate the image quickly andaccurately.The traditional image low-level characteristics will vary depending on thedifferent illumination of image, rotate, etc. So that it can’t guarantee the stability ofretrieval features. To this end, this paper utilization Scale Invariant FeatureTransform (SIFT) to retrieval image. At the same time, this paper uses LocalityPreserving Projection (LPP) algorithm while using SIFT algorithm for imageretrieval to reduce the dimension, in order to improve the matching feature points inreal time by reducing the number of feature points. Enhanced approximate nearestneighbor method is used in the feature point matching links, which adds a secondaryjudgment mechanism when it is matching. If it’s possible match points, the methodwill execute handshake confirmation, which can improve the matching accuracy. Atthe same time，compare experimental results in different lights, rotation, zoom andother situations. Proof the improved SIFT algorithm has a good match resultscompared with other methods and it can be well applied in image retrieval.This paper selects300images as experimental database and extracts featurepoints, respectively. Then, each picture matches the300images and select fiveimages which having the most match points as its image searching results. At thesame time, record the matching time. Through experimental comparison with theoriginal search method, it can be proved that the improved SIFT algorithm canenhance the real-time and matching rate in image retrieval, so it can be well applied in image retrieval to provide a theoretical foundation for future research of imageretrieval technology. |
PDF Full Text Request