Multi-mode 3D Face Recognition System Design And Implementation

In recent years,the 3D face recognition has been the hotspot in the field of pattern recognition.3D face recognition relative to the fingerprint recognition,iris recognition has the congenital advantage,such as recognition system is mainly reflected in the simple and convenient operation,the target is quite small intrusive,even cannot detect the target of target detection,this is can't do other testing way.However,the present stage of the algorithm accuracy compared with other detection method has obvious shortcomings of most of the robustness of the algorithm is not high,only has good effect on the individual database.Compared with the traditional face recognition,due to the 3D face recognition is not dependent on visible light imaging,so can effectively avoid the traditional light sensitive in face recognition of the problem.At the same time,the 3D than 2D face contains more intuitive face information,if the information correct application of the 3D face recognition algorithm will reap higher than 2D face recognition accuracy.In this paper,We introduced several method to enhance the robustness of this algorithm,so that the algorithm robustness is guaranteed.First put forward a step by step filter tip point positioning method,this method is robust,and complexity are acceptable.Part of face recognition algorithm using a local curve characteristic feature of face,which makes the algorithm can tolerate people face some of the loss of data.Introduces the depth of the projection of 3D face,geodesic curvature projection and projection to make full use of 3D face feature,and proved through the experiment using the three face information is necessary.Used in determining phase,the neural network as a decision method,the algorithm calculation same person face features between different face value,which make the training set very sufficient,the training of the network effect is better.Algorithm is divided into preprocessing,face pose correction,characteristics calculation,using neural network to determine four steps.At the end of the paper discussed the shortcomings of 3D face recognition and improve vision.