Accurate image quantization adapted to multisource photometric reconstruction for rough textured surface analysis

In classical photometric stereo (PS), a Lambertian surface is illuminated from three distant light sources to recover one normal direction per pixel of the input image. In continuous noiseless cases, PS allows us to reconstruct the textured surfaces in three-dimensions with a high degree of accuracy and a high resolution. In the real world, an image is a digital quantization, a limited and noisy representation of a surface. In this paper, we present an accurate 3D recovery approach for real textured surfaces based on an acquisition PS method. The proposed method uses a sequence of images for each light source to recover an accurate and unlimited representation of a surface. To evaluate the performances of the proposed method, we compare it to other traditional PS methods on real textured surfaces.