Research And Implementation Of Eye Location Algorithm Based On Face Image In A Mobile Environment

With the popularity of mobile devices,face recognition under uncontrolled environment has received increasing attention.However,in most of the face recognition system,face image needs to be aligned basing on the eye coordinates.Therefore,eye location is an important part of face recognition system,and the accuracy of eye location directly affect the performance of the face recognition system.But the existing eye location algorithm can not adapt to the uncontrolled environment,because of the complex appearance of the eyes.With the enlargement of the face recognition application and the increase requirement of the accuracy,realizing eye location precisely under uncontrolled environment is particularly important.Therefore,combining the characteristics of the uncontrolled environment,the paper established a new model and algorithm for eye location.Besides,we also propose a face correction scheme and design and implementation of face recognition system with the results of eye location.The main work is as follows:(1)To alleviate the problem of large amount of data and slowly speed in SVM training,we proposed an improved two-level SVM algorithm.Firstly,proceeding ascending order to each category,and then choose the front sample from each queue as boundary samples of each category,finally according to the KKT conditions adjust initial setup iteratively throughout the training sample set until convergence.In order to speed up the testing,this paper implemented the two-level of SVM classification.Experiment result shows that the proposed improved two-level SVM algorithm can reduce the amount of training data and improve the training speed,with ensuring the accuracy of classification.(2)For eyes complicated appearance and partial occlusion under uncontrolled environment,we proposed a improved PS(Pictorial Structure)model.Firstly detecting face from image,and make class labels for eyes and nose aiming to shrink PS modeling area.Improve basic PS model constraint method,combining global and local structural constraints to form new inverted trapezoidal constraints which contains two wing of nose and two eyes.We adopt a heuristic method to handle the partial occlusions with eyes.Experimental results show that the proposed improved PS model can achieve higher precision of eye location under uncontrolled environment.(3)Based on improved PS model and improving two-levels SVM classification eye location algorithm,we designed a face posture correction and face recognition scheme.Analyzed the process of face posture correction and the requirements analysis,database design and system design of face recognition system.(4)This paper implemented a face recognition system based on eye location.The experimental results show that the eye location algorithm proposed in this paper can accurately location eye,and can greatly improve the accuracy of face recognition.