Abstract: Aiming at the problem that the correct matching rate of feature points in traditional SIFT (Scale Invariant Feature Transform) registration algorithm is low and the effect of registration is poor, a new adaptive fractional order SIFT algorithm is proposed for image registration. The algorithm first constructs an adaptive fractional-order mathematical model based on the gradient modulus and information entropy of the image, and automatically calculates the best fractional order of each pixel. Secondly, constructs an adaptive fractional differential mask based on the optimal fractional order. And integrate it into the SIFT algorithm to extract more precise and effective key points, thus improving the precision of the SIFT algorithm; In the feature point matching stage of SIFT algorithm, the cosine similarity constraint is added when the similarity measure is performed, which solves the problem that the Euclidean distance cannot determine the spatial positional relationship of the feature vector, and further improves the accuracy of feature point matching. The improved random sample consistency algorithm (RANSAC) is used to further reduce the mismatched feature point pairs. Finally, according to the matching features point pairs calculate the spatial transformation matrix to achieve image registration. The verification results show that the matching accuracy of the algorithm is higher and the quality of registration is also improved significantly.