An improved SIFT algorithm based on adaptive fractional differential

Abstract

Scale invariant feature transform (SIFT) has limitation in extracting features accurately for the images with small gradient and weak texture caused by low contrast. In order to tackle this problems, this paper proposes an improved SIFT algorithm based on adaptive fractional differential. The method firstly construct a mathematical model of adaptive fractional differential based on local image information, therefore, the relationship between the optimal order and image local information can be built up, and the optimal order at every pixel can be calculated automatically according to the characteristics of image. And an adaptive fractional differential dynamic mask is constructed in term of the optimal order and Riemann–Liouville (R–L) fractional definition. And then it is applied to SIFT algorithm for image matching. The method proposed in this paper is an important extension of SIFT algorithm. The theoretical analysis and experiment results indicate the proposed algorithm is capable of matching image with small gradient or weak texture or weak edge.