利用Tikhonov正则化改进移动最小二乘的图像变形算法  

Moving Least Squares Based Image Deformation Algorithm Improved with Tikhonov Regularization

作　　者：崔小曼 于凤芹[1] Cui Xiaoman;Yu Fengqin(School of Internet of Things Engineering,Jiangnan University,Wuxi,Jiangsu 214122,China)

机构地区：[1]江南大学物联网工程学院

出　　处：《激光与光电子学进展》2019年第23期92-98,共7页Laser & Optoelectronics Progress

基　　金：国家自然科学基金(61573168);中央高校基本科研业务费专项资金(JUSRP51733B)

摘　　要：针对移动最小二乘算法在图像变形过程中,求解的线性方程组系数矩阵会出现不可逆、求解不稳定的问题,通过引入Tikhonov正则化,运用L-曲线法求解正则参数,对系数矩阵施加约束项从而得到精确解,避免病态方程组的形成;针对在实现图像变形过程中,定位特征点人工量大且特征点不足的问题,运用Dlib库自动提取68个覆盖人脸五官和轮廓的特征点。仿真实验结果表明,与原算法相比,提出的改进算法可以使图像产生清晰、准确的变形。In the image deformation process of the moving least squares based algorithm,the coefficient matrix of the solved linear equations is irreversible and unstable.In this study,we apply a constraint term to the coefficient matrix to obtain the exact solution and avoid the formation of ill-conditioned equations by introducing Tikhonov regularization and using the L-curve method to solve the regular parameters.To overcome the limitation of a large number of manually localized feature points and an insufficient number of feature points in the process of image deformation,the Dlib library is employed to automatically extract 68 feature points covering facial features and contours.Simulation results demonstrate that,compared to the original algorithm,the proposed algorithm can produce clear and accurate image deformation.

关 键 词：图像处理 图像变形 移动最小二乘 TIKHONOV正则化 Dlib库 

分 类 号：TN919.85[电子电信—通信与信息系统]