Research Of 3D Face Recognition Based On The Local Features

Face recognition has been widely used in electronic commerce,access control system,security defense and other fields because of its advantages such as optional,non-contact,concealment and simple operation.Compared with the traditional 2D face recognition,3D face recognition is based on the 3D data of human face.Although it is not affected by illumination,pose and makeup,it is very sensitive to expression variations.In order to solve the problems caused by expression variations on 3D face recognition and improve the recognition performance,two kinds of 3D face recognition algorithm are proposed in this paper.The main work and innovations are as follows:1)A 3D face recognition algorithm based on profiles and the local descriptor is proposed.Firstly,after preprocessing,many profiles are extracted in the areas less affected by facial expressions according to the facial symmetry plane which are used as key points after sampling.Next,the neighborhood of the key point is constructed using the DAISY descriptor,then histograms of shape indices and slant angles and difference of shape index is used as the the descriptor of the neighborhood circle.The feature of each key point is obtained by cascading the descriptors of all neighborhood circles along the main direction.Finally,the classification is completed by the angles between the features of the key points.Experiments show the features of the semi rigid region can be fully utilized,and the effect of facial expression changes can be weakened.2)A new algorithm for 3D face recognition based on radial curves feature is proposed.Firstly,the face data of 9 scales is obtained by the Gaussian filter.At each scale,18 radial curves of human face are extracted according to the symmetry of human faces.Next,the geometric histogram feature of every profile is extracted,and the feature of the face is obtained by cascading the features of all radial curves in all scales.Then the dimension of the feature vector is reduced.Finally,the distance metric learning algorithm is used to learn the covariance matrix in Mahalanobis distance,and faces are classified by using the k-nearest neighbor classification algorithm.Experiments show the semi-rigid region can be fully utilized,meanwhile the non rigid region information is also utilized effectively,so the recognition rate is higher in this algorithm.