Feature Extraction and Filter Design for Eye Pattern Analysis

Abstract

Abstract This paper introduces alternative solutions to recognize the state of eyes (being open or closed) and detect eye corners. First, the determination of the state is based on a binary dual-state classifier which is trained by boosting the proposed Local Intensity Increasing Patterns (LIIP) based feature vectors extracted from eye samples. LIIP encodes the increasing trend of local intensity pattern; Second, a novel correlation filter, the Synthetic Least Squares Filter (SLSF), is developed to detect eye corners. SLSF reversely constructs the filter to achieve a least squared error between the actual and synthetic correlation responses. Experimental results show the proposed methods can achieve favorable performance.