Fast algorithm for digital retinal image alignment

An algorithm based on an automatic multistage, multiple-featured-based registration strategy has been developed to decrease the amount of computation and to increase the speed of alignment for digital retinal images. The retinal image is digitized and processed to detect the location of the optical disk as a coarse feature. The branch point of a retinal vessel is automatically set up as a template by multilevel vessel tracing with adaptive-thresholding, cavity-filling, neighborhood-averaging, and stem-thinning techniques. Flowcharts for the automatic template setup and image alignment are given.<>