Research On 3D Face Recognition Based On Ricci Curvature Flow

In modern society,the recognition technology based on biometrics such as iris,fingerprint and human face has huge market demand in many fields of society.Face recognition with its natural,friendly and non-touch advantages,has a huge development prospects.Although the existing three-dimensional face recognition algorithms already obtain higher recognition efficiency under certain conditions,they are still limited as the face itself is a non-rigid surface,and if it is affected by light and facial expressions and posture changes during the recognition process.At the same time,the direct calculation on 3D face surface will greatly reduce the efficiency of the algorithm largely due to its large amount of spatial geometry information.Therefore,if the 3D face is reduced to two-dimensional plane,it will cause a lot of loss of geometric information in 3D human face.The main works of this paper are as follows:Aiming at the loss of the geometric information of the three-dimensional surface caused by the existing three-dimensional face recognition algorithm in the process of dimension reduction,by studying the conformal geometry related knowledge,this paper adopts the conformal mapping method based on Ricci curvature flow to reduce the three-dimensional face surfaces to two-dimensional planar disk.The geometric information in the 3D surface is not lost as much as possible in order to improve the recognition efficiency in the subsequent recognition process.A three-dimensional face recognition method based on vertex energy minimal pattern(VEMP)is proposed.Each point on the surface will generate a unique energy value in the process of decreasing the dimension through Ricci curvature flow.By counting the variation of the energy value of each vertex,the feature histogram is generated and used as therecognition basis.Experimental results show that the proposed algorithm has higher recognition rate and robustness.In the view of the impact of non-rigid deformation caused by facial expression changes on the system recognition efficiency,an improved method of concentric circle weighted VEMP(CCVEMP)is proposed.After the dimensionality reduction is completed,the entire face surface is normalized,and the tip of the human face is extracted as the center of the circle.The concentric circles are formed with the size of 40% and 70% of the radius of the entire 2D plane disk,and thus the facial surface is divided into three regions with different weights.Experimental results show that this method has better recognition effect.