A fast affine-invariant features for image stitching under large viewpoint changes

Image alignment and stitching is a popular application on many smart phones, but it is time consuming and creates a critical bottle neck in the course of implementation. In this paper, a fast and high-quality image stitching method is proposed. First, a series of simulated images is obtained by simulating the latitude and longitude angles of a raw image; second, FAST detector is used to detect the features of all the simulated images and described by Fast Retina Key-point (FREAK) before all the feature information is projected to the raw image; third, Hamming distance is used as a feature similarity metric and all the features are matched directly instead of using the repetitive projection in Affine-SIFT (ASIFT). RANSAC is then used to achieve the optimal affine-transformations, and lastly, a weighted average bending algorithm is used to smooth the intensities of the overlapping regions. The experimental results demonstrate that the proposed image stitching method greatly increases the speed of the image alignment process and produces a satisfactory result.