Integration of a Normal Field without Boundary Condition

We show how to use two existing methods of integration of a normal field in the absence of boundary condition, which makes them more realistic. Moreover, we show how perspective can be taken into account, in order to render the 3D-reconstruction more accurate. Finally, the joint use of both these methods of integration allows us to obtain very satisfactory results, from the point of view of CPU time as well as that of the accuracy of the reconstructions. As an application, we use this new combined method of integration of a normal field in the framework of photometric stereo, a technique which aims at computing a normal field to the surface of a scene from several images of this scene illuminated from various directions. The performances of the proposed method are illustrated on synthetic, as well as on real images.