A Comparison of Regularization Methods for Near-Light-Source Perspective Shape-from-Shading

被引：1

作者：

Ahmad, Bilal ^{[1
]}

Floor, Pal Anders ^{[1
]}

Farup, Ivar ^{[1
]}

机构：

[1] Norwegian Univ Sci & Technol, N-2815 Gjovik, Norway

来源：

2022 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING, ICIP | 2022年

关键词：

Shape-from-Shading; near-light-source; anisotropic diffusion; L2; regularization; L1;

D O I：

10.1109/ICIP46576.2022.9897921

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

3D shape reconstruction from images is an active topic in computer vision. Shape-from-Shading is an important approach which requires the surface properties and light source position to infer the 3D shape. A L2 regularizer is typically used to penalize the irradiance equation. In this article, anisotropic diffusion (AD) is introduced as a regularizer to solve the image irradiance equation. The method is then compared with L1 and L2 regularization methods, where all of the three techniques are formulated using gradient descent. Results shows that with AD, edges can be better preserved. AD shows lower depth error and higher correlation when compared with L1 and L2 regularization methods.

引用

页码：3146 / 3150

页数：5

共 13 条

[1]

Bas M., 2021, Information Technology in Biomedicine, V1186, P81, DOI [10.1007/978-3-030-49666-17, DOI 10.1007/978-3-030-49666-17]

[2] Numerical methods for shape-from-shadling: A new survey with benchmarks [J].