Research On 3D Face Recognition System Based On Kinect

With the development of science and technology,face recognition is gradually playing an increasingly important role in society.Two-dimensional face recognition technology tends to be mature but greatly affected by illumination,posture,expression,etc.Because of these limitations,3D face recognition becomes to be the hotspot in face recognition field.However,the professional 3D scanning devices are generally expensive and complicated to operate,and 3D face feature extraction algorithms also cost much time.In order to solve this problems,this paper studies the feasibility of completing 3D face recognition system with low-cost Kinect depth sensor.The main works are as follows.1.Depth image acquired by Kinect is noisy,and cannot be directly used in the face recognition process.The traditional face detecting method by using the information of feature points can't achieve high detection efficiency.Aiming at this problem,we use the advantages of Kinect's ability of obtaining RGB-D images,combine RGB information and depth information to detect face region.Firstly,we detected face by Adaboost and skin color criterion algorithm on RGB image,and mapped the result to aligned depth image.Secondly,filtering the face out which in invalid range of Kinect,the depth information of the face region can be finally extracted.2.The 3DLBP description operator's length is too large and cumbersome to effectively reflect regional identification characteristics.Based on 3DLBP algorithm,a feature extraction method combining central symmetry 3DLBP features and central LBP features is proposed.Firstly,the central symmetric 3DLBP operator is applied to extract feature of the depth image.Secondly,in order to compensate for the lack of depth features of the central pixel,uniform LBP operator is added.We use two public datasets and laboratory self-built dataset to verify the validity of improved operator,the experimental results show that compared with 3DLBP,the accuracy of 0.34%,1.20% and 1.39% was improved,and the feature extraction time was also shortened by 24.09%,24.18% and 24.67%.3.With the previous research,an automatic face recognition system based on Kinect depth information is designed.The complete process from face information registration to face recognition is completed,which verifies the effectiveness of the proposed method and the feasibility of 3D face recognition with low cost Kinect equipment.