Introduction - If you have any usage issues, please Google them yourself
SIFT by the LDA, and then binarization formation of short features, in order to improve the matching speed, reduce the amount of storage the characteristics of the Low, False, the Positive, the better performance than SIFT, but need training.