基于各向异性全变分的GBR参数估计

Estimation of GBR Parameters Based on Anisotropic Total Variation

GBR参数估计是求解未标定光度立体技术的一般方法。本文针对GBR参数估计中模糊深度不连续的问题,引入具有保持深度不连续的各向异性扩散全变分做正则建立新的变分模型,并运用具有超线性收敛的拟牛顿法对模型进行求解。实验表明各向异性扩散全变分方法能有效降低法线误差,提高深度图不连续,且拟牛顿法求解该模型所用时间远低于EM方法和DM方法。

GBR parameter estimation is a general method for solving uncalibrated photometric stereoscopic technique. In order to solve the problem of fuzzy depth discontinuity in GBR parameter estimation, a new variational model is established by introducing the anisotropic diffusion total variational model with depth discontinuity, and the quasi-Newton method with superlinear convergence is used to solve the model. Experiments show that the anisotropic diffusion total variational method can effectively reduce the normal error and improve the discontinuity of depth map, and the time of quasi-Newton method to solve the model is much lower than that of EM method and DM method.