Research Of 3D Face Recognition Under Partial Occlusion And Expression Variations

Face recognition technology has extensive applications in many fields because of its friendliness,nature and intuitive.After decades of development,the 2D face recognition technology based on image information has achieved great success,but it is still difficult to solve the problems that caused by illumination,posture and makeup.Because 3D face recognition technology can make full use of spatial geometric information,so it is not sensitive to these problems.However,3D face recognition also faces severe challenges such as partial occlusion and expression changes.In this paper,some studies have been done based on the analysis and summary of the existing 3D face recognition algorithm.The main work and innovation are as follows:1)A 3D face recognition algorithm based on occlusion detection and local features is proposed.First,the non-occluded average face is trained,then the nasal tip of the average face is detected,and the face is divided into multiple subspaces by offsetting with the nasal tip.At the time of occlusion detection,the points with smaller shape index and higher curvature are first positioned,and these points and their composed regions are regard as occlusion candidate regions.Then calculate the difference between the face and average face about these regions to determine the final occlusion region.In the face recognition stage,the face keypoints are obtained by sampling the radial strings,and then the 3DWWs(3D Weighted Walkthroughs)are constructed as local descriptors.Finally,the multi-task sparse representation is used to completes the classification.The experimental results on FRGC v2.0 database and Bosphorus database fully prove the effectiveness of the algorithm and robustness to partial occlusion.2)A 3D face recognition under partial occlusion and expression variations using radial strings is proposed.First,radial strings are obtained by rotating around the nasal tip with the mirror plane of symmetry,and the radial strings in the semi-rigid region of the face are selected as an effective part,then intercepting and sampling in the effective part of each radial string.Taking the distances of two adjacent sampling points as diameters,the center point of the two adjacent sampling points as the center draws a ball to obtain the local region.Extracting three local features constitutes a local feature descriptor.Finally,the local features are used to construct the cost function and generate the similar vectors.The similarity vector divides the radial string into multiple segments to avoid irregular deformation regions caused by local occlusion and expression variations.Finally,the radial strings are matched by the similarity vector,then the three-dimensional face recognition with partial occlusion and expression variations is achieved.The experimental results in the FRGC v2.0 database and the Bosphorus database fully prove the effectiveness of the algorithm and robustness to partial occlusion and expression variations.