Fast automatic fundus images registration and mosaic based on Compute Unified Device Architecture

Abstract

In order to overcome the characteristics of low contrast, non-uniform illumination, limited field of view (FOV), and the geometric distortion between different FOV of the fundus images, a fast automatic fundus image registration and mosaic algorithm based on Compute Unified Device Architecture (CUDA) is presented. Firstly fundus images are enhanced by homomorphism filtering, then the Scale Invariant Feature Transform (SIFT) features in effective FOV are extracted and matched between images with CUDA speeded up. With CUDA application, point pairs are purified using random sample consensus (RANSAC) algorithm employed perspective model, transformation matrixes are computed according to the matching point pairs of surrounding FOV images to the central, image registration and image fusion is implemented to get fundus panoramic image finally. The automatic registration and mosaic results of multiple FOV images obtained by fundus camera show that the algorithm is robust and stability with registration accuracy up to pixel level, the algorithm speed upgrade 10 to 30 times, high-precision automatic fundus image mosaic can be achieved.