3D Face Recognition By Fusing Surface Shape And Texture Features

In the past decade,there has been a lot of 3D face recognition research used for identification.Add the face intrinsic characteristics of shape and texture information,3D face images with universality and non-invasive,has now become a very promising research direction,it solved the limitations of 2D face,such as posture,facial expression,illumination changes and the impact of its shade.And initial research of 3D face is colour image and depth image,but it is not really a 3D image,to some extent,lost a part of the information such as geometry.And then there are some ways of projected 3D images onto a 2D image,adopt the method of 2D research method of face recognition.With the development of the latest swift and violent development of computer hardware,as well as 3D scanning,the progress of technology,to promote the development of 3D face recognition,improve recognition accuracy and the reliability of the application.In this context,the research of SVM method based on Mesh-LBP fusion is carried out.And obtained the following research results;Improved adaptive face segmentation method is proposed,the fusion of many kinds of 3D geometry and texture information,and the brightness information feature extraction method,the combination of support vector machine SVM classification prediction methods.The main research contents of thesis are as follows:(1)For 2D face recognition is easy to be affected by illumination,pose change and self-occlusion,3D face contain more information.This thesis researches 3D face recognition method by fusing surface shape and texture features.The weight function of the surface has Gaussian curvature,mean curvature,maximum curvature,minimum curvature,gray value,covariance and shape index,researches the scheme of different descriptor.(2)For general 3D face segmentation method to tip as the face center,with a certain length to radius to face segmentation method of disadvantages:easily affected by environmental changes and its expression,may appear divided seriously absent or contains the collar,etc.The problem of redundant information.In this thesis,we study the adaptive modified face segmentation method.To contain the head below the redundant information for simple pretreatment,remove redundant information,such as clothing and for nasal fip point,at its tip radius of the distance to the top of your head multiplied by a certain weight of the ball to split out the 3D face.A large number of experiments show that this method is significantly improved than the original method.(3)Through the front of a variety of 3D face feature information extraction and fuses in together,histogram normalization,in combination with support vector machine(SVM)for training the SVM classification,and prediction of the final recognition result.SVM mathematical theory perfect,the requirements for the training sample is not much,only need a small number of support vector can be obtained by decision function,has strong generalization ability,and make the global optimal solution,widely used in pattern recognition and computer vision,etc.The experimental results show that SVM has higher accuracy than traditional methods after three to four rounds of relevant feedback.