Research On 3D Face Recognition Based On Frenet Geometric Feature

Many 3D face recognition algorithms emerged in recent years.The application of 3D data in face recognition has great advantages.It contains facial spatial geometric information,which is less affected by illumination,expression and posture than twodimensional data in face recognition.The 3D face recognition is becoming a popular research direction.However,due to the improvement in the data dimension,the complexity of calculation of 3D data will also increase.The methods of manually selecting feature points or three-dimensional space projection to a two-dimensional space are usually used to reduce the computation amount,but these methods could not make full use of the advantages of three-dimensional data.Therefore,this thesis proposes the Frenet geometric feature for 3D face recognition.and the acquisition of this feature does not require redundant operations such as 2D projection,and directly processes the 3D face data.The specific research contents of this thesis are as follows:Firstly,after preprocessing the 3D face data,two kinds of spatial curve representation of the 3D faces are compared,and the radial curve with better efficiency and robustness is selected as the spatial curve to be extracted,which can greatly weak the effect of expression on 3D face recognition.Afterwards,the Frenet formula proposed in this thesis is derived to obtain four different Frenet geometric features,and the pose invariance of the four features is demonstrated.Four geometric features of the 3D face radial curve are extracted for classification tasks.Finally,three methods are used to reduce the geometric features dimension,and the final recognition results are obtained by inputting the minimum distance classifier.In the stage of evaluating the experimental algorithm,this thesis compares the recognition rates by using different dimensionality reduction methods,then compares the recognition and computation speed of different algorithms to further prove that the proposed method has more competitive recognition performance and higher computational efficiency.