Attribute reduction for SIFT local descriptors using PCA and CAIM

Stable local feature detection and representation are the fundamental components of target recognition and image retrieval. The traditional SIFT algorithm's descriptor of the feature points is a 128-element vector, and a lot of redundant information is presence. So the brief and effective expression of the image feature information is the key to improve the performance of the algorithm. This paper is to use PCA and CAIM method to extract the more concise and more robust descriptor. Image feature points matching experiments show that the improved algorithm has a higher matching accuracy and faster matching speed.