Research On 3D Face Recognition Based On Local Descriptor

The 3D face recognition attracts more and more attention because of its insensitivity to the variance of illumination and pose.Unfortunately,expressions take severe problems for 3D face recognition.So it is crucial for 3D face recognition that how to extract expression-robust features.This paper mainly studies the 3D face feature extraction based on local descriptors and recognition algorithms.The main work and progress are as follows:1.3D face preprocessing.The nose tip is automatically obtained by calculating shape index.Then a sphere with radius of appropriate length and center at nose tip is used to extract the main area of a 3D face.Sharp areas are removed based on Euclidean distance between points and the pose of 3D face is adjusted using Principal Component Analysis.Experiments show that the above preprocessing can automatically obtain the valid area of 3D face recognition and provide high quality data for the subsequent feature extraction.2.Extraction of expression-robust features.To reduce the effect of expression,a 3D face is divided into 11 partitions.For each partition,the edge detection method of point cloud based on 2D projection is applied to detect the non-edge points.Then the normal,depth and shape index information of each non-edge point is extracted and encoded by 3D-LBP to construct its local descriptor.The histogram is utilized on every partition to reduce the dimension of facial features and acquire the feature vector of each partition.3.3D face recognition.In order to verify the validity of the local-descriptor based feature for 3D face recognition,Partition-Weight KNN and SVM matching strategies are used in this paper.The recognition rate of experiments on GavabDB 3D face database reach to 94.50%and 93.15%.Experiment results show that it is feasible to apply the local-descriptor based method on 3D face recognition under expression varieties.4.A prototype system of 3D face recognition based on local descriptor is designed and implemented.The system integrates the face data preprocessing,feature extraction and face recognition algorithm proposed in this paper.The input face is processed by a series steps and then a classification number is given.