Surface Smoothing for Enhancement of 3D Data Using Curvature-Based Adaptive Regularization

This paper presents both standard and adaptive versions of regularized surface smoothing algorithms for 3D image enhancement. We incorporated both area decreasing flow and the median constraint as multiple regularization functionals. The corresponding regularization parameters adaptively changes according to the local curvature value. The combination of area decreasing flow and the median constraint can efficiently remove various types of noise, such as Gaussian, impulsive, or mixed types. The adaptive version of the proposed regularized smoothing algorithm changes regularization parameters based on local curvature for preserving local edges and creases that reflects important surface information in 3D data. In addition to the theoretical expansion, experimental results show that the proposed algorithms can significantly enhance 3D data acquired by both laser range sensors and disparity maps from stereo images.