抄録
This paper presents a robust method for eye shape detection. Existence of eye components such as pupil, iris, eyelid and sclera makes the color variation between the eye and skin very sharp. This variation is measured by applying Principal Component Analysis (PCA) on RGB color channels of an eye image. In addition, eyes have an elliptical structure which can easily be detected in Logpolar domain. Log-polar transform (LPT) decomposes the eye into two parabolas along θaxis. Thus, the eye shape can be detected by searching on corresponding coordinates along log r axis. Based on these properties of eye representation by PCA and LPT, the eye shape is extracted regardless the change of scale, rotation and lighting conditions. The results show high accuracy with reducing time consumption compared to the other existing methods.
