A Synthetic Dataset for Deep Learning Recognition of Pathological Iris Affected by Coloboma

Abstract

Biometric recognition systems might not work for people suffering from alteration of physical characteristics. This can also happen for well-known iris recognition systems. In this paper, we describe the creation of a synthetic dataset of eyes suffering from Coloboma, a congenital abnormality of eye membranes characterized by a “keyhole” appearance of the pupil. Due to the rarity of the disease, we apply image processing techniques on a dataset of healthy eyes to artificially simulate the effects of Coloboma. The pupil is distorted to simulate Coloboma on each of these images and the iris is compressed in the direction of the defect. A preliminary evaluation based on k-means has been performed. The dataset will be adopted for designing “non-excluding” iris recognition systems.