A Neural-Learning-Algorithm-Based Shape from Shading System

This paper introduces a shape from shading (SFS) system, which is based on a neural learning algorithm. The system takes a single image of an object with a CCD camera and then reconstructs the object surface with a neural-learning-based SFS algorithm. This SFS algorithm solves and optimizes the neural elements, named network weights, by minimizing the cost function that is composed of the intensity constraint and the integrability constraint. This neural-learning-based SFS algorithm can provide a promising effectiveness and accuracy. Moreover, the reconstructed surface of this system can be applied in the area of surface measurement and defect detection