Open/closed eye recognition by local binary increasing intensity patterns

Open/closed eye recognition is an open topic in many intelligent systems. This paper proposes a novel appearance-based method to solve the problem. It is considered as a two-class classification task, and two important components are: feature descriptor and classifier learning. Originally, this work introduces a distinct local feature named Local Binary Increasing Intensity Patterns (LBIIP), which uses one decimal label to specify the intensity increasing tendency of the local region around each pixel. It inherits the merits of both Local Binary Patterns (LBP) and gradient features. Given an eye image, numerous sub-windows are obtained by scanning at various scales and locations. Then the LBIIP-Histograms (LBIIPHs) are extracted from the sub-windows, and concatenated into a feature vector (descriptor). On the other hand, Discrete AdaBoost is applied to selecte a few most discriminative features and learn the open/closed eye classifier. Experimental results show that the proposed approach is very fast and efficient.