A speed-up and robust image registration algorithm based on FAST

Abstract

A speed-up and robust image registration algorithm for multimodality images is proposed in this paper. The median filter is used to suppress the noises in the reference and distorted image. Then an improved FAST (Features from Accelerated Segment Test) method with a robust threshold is implemented for corner detection. In order to match the feature points, a point matching approach based on threshold learning is proposed in this paper. Besides, the pair of feature points is matched in regional area according to the gray correlation. After calculating the affine parameters, the distorted is registrated. This scheme is tested by three different modalities real images including nature images, space images and infrared images. Results reveal that the proposed image registration algorithm can be finished within 20ms for multimodality images.