An improved scale invariant feature transform algorithm
Jianfang Wu · Harbin Gongcheng Daxue Xuebao/Journal of Harbin Engineering University · 2010
The scale-invariant feature transform(SIFT)algorithm,as applied to image stitching,often selects feature points that have no value.This causes poor real-time performance.To overcome the problem,the process of searching extremum in the whole scale space was explored and then a geometric theory of invariant moments was used instead.Image edge extraction was then limited to the edge group of the image.Extraction of feature points in the scale space corresponding to the edge class yielded an improved SIFT algorithm.The results of experiments showed that the improved SIFT algorithm reduced feature points by 20% ～50%.This greatly reduced SIFT feature point redundancy and increased the speed of the algorithm.