Study On The Algorithms Of Shape-from-shading

The research on 3D surface shape reconstruction is an important field in computer vision. At present, many reconstruction techniques have been developed. The method of shape-from-shading (SFS) makes use of the change of image grey values to reconstruct 3-Dparameters such as the relative height and surface normal. SFS is one of the key techniques of 3D shape reconstriton in computer vision. This method has been intestagted widely because it is not only easily realized but also applicable for the condition in which other methods are not suitable.In present, most of the reconstruction SFS methds were based on the Lambertain reflectance model. But the Lambertain model can not describe the characteristic of speculare reflectance. Popular reflectance models used in SFS methods were analysized firstly. And experiments on synthetic images showed that hybrid model described reflectance characteristics of specular reflectance more accurately than Lambertian model. Reconstruction error may accumulate along diagonal directions using forward or backward differential. So the reflection map equation of hybrid reflectance model was discereted using central finite difference approximation method, and algebra equation was arrived. Then Newton iteration algorithm and general iteration algorithm and were used to solve the equation and 3-D surface heights were acquired. Experiments showed that central difference algorithms may tide over the accumulated error of forward or backward differential, and the average Newton iteration algorithm improved the accuracy of reconstruction. At last,3-D surface of the Moon was investagted using central finite difference approximation method based on Lommel-Seelige reflectance model. Both the imaging conditions and algorithm were given. And the experiments on synthetic images and real image of the Moon were performed.