A comparison of different features for automatic eye blinking detection with an application to analysis of deceptive behavior

Systems for robust and real-time detection and measuring of the duration of eye blinking have been widely used in many different applications. Initially, eye blinks were used to provide an input modality to people with severe disabilities so as to be able to access Human Computer Interaction (HCI) systems. Recently, eye-blinking behavior was used as an indicator of deceptive behavior. In this paper we present a comparison of different features such as raw-image intensities, the magnitude of the responses of Gabor filters, Histograms of Oriented Gradients (HOGs) and optical flow extracted to be used for automatic eye blinking detection. For the classification of the above mentioned features we employed Support Vector Machines (SVMs) with Gaussian Radial Basis Function (RBF) kernels. We assessed their performance using eye samples from a controversial TV show recently aired, the so-called Moment of Truth. In the end, we used the eye-blinking algorithms that provided the best results in order to analyze the eye-blinking behavior in cases in which the player answered truthfully and in cases he/she lied.