Automatic pupil detection on retro-illumination lens images from a population-based study

Automatic detection of the pupil is an important step in cataract assessment. This paper proposes a new pupil detection method which addresses the challenges faced in current methods employed for this task. We evaluate the performance of the proposed method against the current methods on a large population-based image dataset of more than 9000 images from the Singapore Malay Eye Study (SiMES) database. The accuracy achieved is 98.60% for the proposed method for SiMES-1. A modified version of the method was applied on the follow-up SiMES-2 dataset, obtaining an accuracy of 97.25%. The results are encouraging towards a fully automatic cataracts detection and assessment.