Iris center localization using integral projection and gradients

Iris center localization becomes more and more popular in the applications of face alignment, human-computer interaction and gaze estimation. However, techniques for low-resolution and low-contrast eye images which are taken in uncontrolled circumstance still need improvement. This paper presents an accurate and fast algorithm to realize iris center localization, by combining integral projection and image gradients. First, the vertical integral projection function is adopted to get precise boundaries of the eyes and to generate a weight map. Second, a novel method combining dot products of gradients with the weight map is proposed. The iris center is detected by searching the maximum of multiplication between the dot products and the weight from the map. A commonly used evaluation is applied to test our result on the very challenging BioID database for eye center and iris center localization. Experiments show that this method can accurately locate irises under natural light conditions and has characteristic of low computation complexity.