A neural-learning-based reflectance model for 3-D shape reconstruction

• Published in: IEEE transactions on industrial electronics (1982) Vol. 47; no. 6; pp. 1346 - 1350
• Main Authors: Siu-Yeung Cho, Chow, T.W.S.
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.12.2000; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Computer vision; Illumination; Image reconstruction; Learning; Light sources; Lighting; Manufacturing industries; Mathematical models; Neural networks; Optimization methods; Reflectance; Reflectivity; Robustness; Shape; Surface reconstruction; Variational methods
• ISSN: 0278-0046; 1557-9948
• DOI: 10.1109/41.887964

In this letter, the limitation of the conventional Lambertian reflectance model is addressed and a new neural-based reflectance model is proposed of which the physical parameters of the reflectivity under different lighting conditions are interpreted by the neural network behavior of the nonlinear input-output mapping. The idea of this method is to optimize a proper reflectance model by a neural learning algorithm and to recover the object surface by a simple shape-from-shading (SFS) variational method with this neural-based model. A unified computational scheme is proposed to yield the best SFS solution. This SFS technique has become more robust for most objects, even when the lighting conditions are uncertain.