说明：  SIFT经LDA，然后二值化形成简短的特征，以提高匹配速度、减少存储量；该特征在Low False Positive 段，性能比SIFT好，但需要训练。
(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.)