Automatic Localization of Trackable Features in Retinal Fundus Images

Laser photocoagulation is used to reduce vision loss due to degenerative eye conditions by guiding a treatment laser over diseased tissue on the retina. However, due to natural motions of the eye healthy retinal tissue can be damaged by the laser. Compensating for this motion can reduce this risk. Motion analysis of scanning laser opthalmoscope (SLO) video provides information suitable for automatic laser aiming. Motion estimation from SLO video can be achieved using landmark-based motion tracking. In the ALET implementation, landmarks are manually selected from a reference image. This thesis work automates the landmark selection process in two steps: first, a set of candidate landmarks is found using a corner detection procedure on a vessel-enhanced retinal image; and second, the set of candidate landmarks are ranked in decreasing order of their estimated tracking integrity.