The Parameters Of Light And Shade Of The Wavelet Neural Network To Restore The Shape Of The Problem

3-D information extraction, as an important aspect of Computer Vision, has been applied in many fields such as nation defence, industry, agriculture, and so on, and has developed with its theory and application continuously. Shape from Shading (SFS), as a method of 3-D information extraction, needs gray images to get the correspond 3-D information of the objects in image, and the image can be obtained without strict conditions. By now, there are mainly three kinds of algorithms in Shade from Shading, namely, directly solving partial differential equations, variational principle and optical method.The method based on Neural Network for Shape from Shading, is a optical method. Thought it need not calculate the environment parameters, it needs certain images in the same circumstances and learns the corresponding surface height. In this paper, a new method is proposed based on Wavelet Neural Network for Shape from Shading with simple image. The main parts in the paper are as following.1. A new error function with adaptable smooth property is presented according to the disadvantage of the uniform control of the shape in optical method, which makes the corresponding changes of the shape and image.2. We construct a especial network based on three Wavelet Neural Network and a nonlinear learning method, give its topological structure, and present a optical method with Wavelet Neural network for SFS problem. The algorithm is described in details and the convergence of the algorithm is presented.3. The uniqueness in SFS problem is argued under conditions of knowing light direction and Lambertian surface assumption.4. The efficiency of the method is verified by examples of both synthetic and real images.