Research And Implementation Of Face Recognition Method Based On Random Forests And LBP Features

As the most promising technology in biometric recognition,face recognition has been widely used in intelligent surveillance,security,finance and other fields.In this paper,face detection,key point location and single sample matching in face recognition system are studied.The results show that the improved random forest model can effectively improve the detection rate and recognition rate of face recognition system.The main works of this paper are summarized as following:1.The gradient Boosting algorithm is used to improve the random forest classification model.Based on this model,an "improved random forest +Haar(haar-like)feature+caseade detection" face detection method is designed.The experimental results on MIT-CMU face database show that the proposed method can improve the average detection rate by 5%-10%compared with the traditional V-J algorithm.2.Based on the regression random forest model,we propose a facial keypoints localization method which uses local face image patches to joint vote on multiple keypoints,and the keypoints localization problem is transformed into a regression prediction problem in voting space.Experimental results on LFW face database show that the proposed method has the highest localization accuracy at 10 pre-defined keypoints compared with the methods based on feature points and shapes.3.To solve the problem of single sample matching recognition,we improve the traditional LBP(Local Binary Pattern)features by using weighted local patches,which makes different local regions have different classification weights.Finally,all local LBP features are weighted fusion,and the matching function is used for face matching recognition.Experimental results on ORL face database show that the recognition rate of the proposed method is 8%-13%higher than that of the traditional method.Finally,based on the three key points,combined with data acquisition and data management module,a complete face recognition system is designed.Through the test in real scenario,the face recognition system designed in this paper has the characteristics of high recognition rate,real-time and strong stability,which can meet the performance and execution time requirements in the real scenario.