3D Face Recognition Based On The Fusion Of Contour And Curvature Information

With the increasing importance of identity verification,face recognition technology is becoming more and more mature.Due to the limitations of the development of twodimensional face recognition technology,in recent years,three-dimensional face recognition technology has been favored by researchers.However,the performance of three-dimensional face recognition algorithm is greatly affected by the quality of database,face posture and expression,and the development of the algorithm is facing great challenges.In this paper,three-dimensional face recognition has been improved in the following aspects:1.In the pretreatment of 3-D face model database,this paper adopts CASIA 3-D database,which was taken by CAS Automation Institute.Because the original data format of the database can not be directly used in experiments,in the database pretreatment module,different from the pretreatment process in other algorithms,this paper uses integrated software to segment and reduce noise,which simplifies the pretreatment process.The model is degenerated to the original point cloud model for experiment.2.In the feature extraction module,in order to improve the accuracy of feature point extraction and better follow-up recognition steps,curvature information and contour information are fused for feature point location in this paper.While locating more feature points to the greatest extent,the accuracy of feature point location is improved,and the inaccuracy of feature point location caused by lack of information in other algorithms is avoided.In contrast experiments,the experimental results show that the proposed feature point location algorithm has certain advantages in both the number of feature points and the location time.3.In the face matching module,two improvements have been made.Firstly,in the construction of feature point descriptors,this paper extends the superior 2D_DAISY descriptor to three-dimensional space,namely 3D_DAISY descriptor.The advantage of this descriptor is that it can integrate the structural relationship between feature points and their neighborhood points to the greatest extent on the basis of maintaining the invariance of attitude and expression.Secondly,in the face matching stage,in order to measure the similarity between the face to be measured and the face in the database,this paper uses the expansion of Hausdorff distance,namely the average Houston distance.The average Hausdorff distance can not only effectively avoid the influence of bad points,holes and noise points on the performance of the algorithm,but also effectively reduce the computational complexity of the algorithm.In the performance evaluation of the algorithm,the neutral expression and neutral attitude parts of the database are used as the database model,and the part with the expression and attitude is used as the database to be tested for recognition experiments.Through analysis,the accuracy of this algorithm can reach 93.5%,and the recognition time of a single feature point is not more than 2.3 ms.The recognition performance of this algorithm is improved compared with other algorithms.