Performance evaluation of local descriptors for face recognition

Abstract

Face recognition (FR) is one of prominent feature in biometric. There are three main steps in FR: face detection, feature extraction and feature matching. Among three steps, feature extraction is considered important as highly distinctive representation will be constructed using certain properties. Researchers have developed several local feature descriptors. However, such those features are proposed to perform well for generalized images. In case of face images which possess distinct feature, it is necessary investigating the effectiveness of such local descriptors to achieve the best possible recognition performance. In this thesis, performance evaluation of local feature descriptors for face images is conducted. Several state of the art local descriptors including Scale Invariant Feature Transform (SIFT), Speed Up Robust Features (SURF), Binary Robust Independent Elementary Features (BRIEF), Binary Robust Invariant Scalable Keypoints (BRISK) and the Fast Retina Keypoint (FREAK), are investigated. We investigate the performance using transformed images with the following properties: scale, blur, rotation, brightness, pose. The performance parameters including repeatability, precision, recognition rate, and computational time are used as measurements. The results indicate that each local descriptor is considered effective in extracting features in different scenarios. Thus, the consideration of choosing local descriptor should take the characteristic of scenario and environment into account.