Research And Implementation Of Improved Iterative Closest Point Algorithm In Fast 3D Face Recognition

Because it is fast,convenient and simple,face recognition is widely used in security filed such as video surveillance,attendance system and personal application fields.With the use of the face as a biometric,the face recognition becomes one of the most attractive research areas in biometrical recognition field and computer vision field.2D face recognition technique has obtained some ideal achivenments,but the 2D face recognition cannot resist the ill effect of the bad factors such expression,lighting and so on,because the 2D images cannot contain enough information.Thanks to the development of the 3D data acquisition and the optimization of the hareware devices,the 3D face recognition technique has been widely researched.The 3D image contains more abundant facial geometric information.Further,the 3D recogniton cannot be interfered by light,posture,makeup and other factors.Therefore,3D face recognition has a broader development prospects.However,the 3D face data is susceptible to the facial expression and collecting those data causes high compuational expense.Therefore,it is very important to find a 3D face recognition model that is insensitive to the expression and is low computational comsuption.This thesis presents a new 3D face recognition model based on the face feature of the rigid region and the improved ICP algorithm.The proposed model is evaluated by the CASIA 3D database.The experimental results show that the proposed model is robust against the ill effect of expression and the computational cost is low.Moreover,it obtains good performance compared with other conventional model.The main work and innovation are as follows:(1)A valuable nasal side contour extraction algorithm based on mean curvature is proposed.By calculating the average curvature of each point on the side contour line,the reference points are precisely positioned to accurately intercept the valuable side contours.Then uses the improved Hausdorff distance to exclude the wrong faces,reduces the unnecessary computation in the face recognition process,and improves the real-time performance of the algorithm.(2)An optimal equal geometric distance contour extraction algorithm based on fuzzy clustering is proposed,and a new algorithm for curve clustering effect evaluation is proposed: The extracted contours are fuzzy clustered,using interclass differences to make sure that leaving a minimum number of contour lines to characterize the most comprehensive face information.The corresponding clustering effect evaluation algorithm is put forward,so as to ensure the optimality of clustering results.(3)Two methods are proposed to improve the ICP algorithm: The calculation parameters of the rigid transformation update quantity in the iterative registration process of ICP algorithm are improved,thus accelerating the global convergence rate of the ICP algorithm and,to a certain extent,enhanced the global optimality of the convergence results of the ICP algorithm;Based on the idea of momentum algorithm,a method of calculating the rigid transformation is proposed,which makes the ICP algorithm more smoothly and quickly in the iterative process to the minimum value of the distance function.(4)A shared photo management tool based on the 3D face recognition framework is implemented.The recognition accuracy and real-time performance of 3D face recognition system are verified.