Research On Reconstructing B-spline Surface From Gray Image

Shape recovery from a gray image is a classic problem in computer vision. In thisthesis, we present a new method to recover the 3-dimension model with a B-splinesurface directly from a gray image. First, the contour of the object is determined, andthe object is segmented from the background; the pixels in the object image aretriangulated, and the dominant points in the contour are chosen by critical pointdetection method to identify the four boundaries of the object image. And then, theimage mesh is mapped into the unit square parameter domain. Second, a B-splinesurface with unknown control points as variables is employed to express the objectmodel, and the intensities of the pixel points are calculated by the Lambertian reflectmodel to construct the energy function. Third, the equations are solved by theconjugate gradient method to determine the control points of the surface. The methoduses the control points of the B-spline surface as variables to reduce the amount ofvariables evidently and improve the efficiency. Moreover, object models withdifferent precision can be obtained by the method. According to the character of theB-spline surface, the so obtained results are smoother than before. In conclusion,since the result is the 3-dimension model of the object itself and can be used in theindustry directly, this method would be used widely in the future.