目前的图像检索技术主要利用图像的颜色、纹理、形状等特征来进行，其检索速度和精确度还不能满足用户需求。采用基于尺度不变特征变换(SIFT)的图像检索，但由于特征点数及维数太大，给检索的实时性造成了影响。对SIFT算法利用局部保持投影(LPP)的方法进行降维，以减少特征点的个数,并利用增强型近似最近邻方法，在匹配时加入了二次判定机制，对可能匹配的点对进行握手确认，从而可以提高匹配的精确度。通过图像库中20幅图像的实验验证，证明了改进的SIFT算法在图像检索中的实时性及匹配率的提高，可以很好地应用在图像检索中。

Currently, image retrieval is based on images color, texture, shape and other characteristics to match. The speed and accuracy of retrieval can not meet the needs of users. Image retrieval based on scale invariant feature transform (SIFT) is carried on. But there are too many feature points and dimensions, which has an impact on real-time retrieve. Locality preserving projections (LPP) in SIFT is used to reduce dimension in order to reduce number of feature points. The enhanced approximate nearest neighbor method is used to improve the accuracy of the match. A secondary judgment mechanism is added, when it is matching. If they are possible match points, handshake confirmation is executed. Experimental results show that based on the experimental verification of 20 images in image library, the improved SIFT algorithm improves the timeliness and matching rate of image retrieval. So that, it can be well applied in image retrieval.