陰影情報およびB-Spline曲面近似による立体形状の認識  [in Japanese] 3D Recognition from Shading and B-Spline Surface Approximation  [in Japanese]

The conventional methods to estimate the shape from shading image data are based on the approximation that a local surface is considered to be a spherical. In these process, it is necessary to calculate the derivative of the shading, and then will be easily affected by the noise in actual use. Therefore, we propose a new method to recognize 3 dimensional shape from shading and B-Spline surface approximation. Using the steepest descent method to minimize the differences between shading data from approximated surface and input image data in each point, we can obtain a free-form surface by B-Spline expression. From the practices applied our method, it is confirmed that we can reduce the effect of noise and reconstruct a 3D shape more precisely than that by conventional methods. As a result, the proposed method is available to recognize a shape from image shading.