Algorithms Of Eye Location In Face Recognition System

With the rapid progress of the information technology, the need for information society to personal identity authentication and recognition is growing. The advantages of face recognition technology are non-contact and convenient. It has become an important research topic of applied mathematics and pattern recognition.In the face recognition system, selecting the appropriate feature for locating has an important influence on the normalization of the image. The eye is a kind of relatively obvious and stable feature on the person’s face with more details, so many scholars have already committed to eyes location. However, eye location has been a difficult issue in the face recognition system, which is susceptible to the influence of facial expressions, posture and other factors. Therefore, the practical eye location algorithm is a very challenging problem.The goal of this paper is to build an eyes positioning module of face recognition system. The main works of this paper is described as follows:Firstly, AdaBoost algorithm and random forest algorithm based on Haar features are studied, and the symmetry classifier is got by means of the mirror symmetry of features. Meanwhile, combined with AdaBoost of OpenCV, they are used in the part of detection and discrimination, which is helpful for enhancing the speed and accuracy significantly.Secondly, using Gabor and template matching method to supplement the location of initial positioning, and using the face symmetry axis to distinguish and update the initial positioning;Thirdly, based on the thought of "first judgment, again classification, from coarse to fine", this pater puts forward a method which can identify the glasses types and distinguish the reflective glasses, analyses the eyes state, and then uses gray characteristics and geometry information of eyes for accurate positioning;Finally, the algorithms proposed have been tested in the CAS-PEAL-R1database, and the accuracy rate is99.43%.In a word, the eye location module developed in this paper has a good, stable performance and has successfully applied to the face recognition system developed by the Applied Mathematics Laboratory of Northeastern University.