Abstract: In view of the fact that the scale invariant feature transform (SIFT) algorithm matching with the mine shaft image does not meet the real-time requirements，an improved SIFT（scale invariant feature transform）algorithm is proposed.In case of keeping the registration accuracy，the algorithm execution time was greatly reduced and the registration accuracy was improved through the reduced scale transformation，simple feature descriptors and considering the Bray Curtis distance as the similarity measure.Experimental results show that the improved SIFT algorithm keeps the linear of the image rotation，the scale and the brightness be changeless，and increases the operating speed at least 4 times higher than that by the original SIFT，resulting in the mineshaft image displaying a good real-time image matching performance.

Key words: Mineshaft image, SIFT, Feature descriptor, Similarity measurement