Combination method: Photometric stereo with Shadows

This paper presents an algorithm allowing to perform photometric stereo with high accuracy results in the presence of shadows. First, we choose a combination of three images suited to reveal the presence of shadow and use it to calculate the normal vector on each pixel. This normal vector is then used to recalculate the intensity in order to evaluate the distance with the real intensities and detect the shadow pixels on each image. If the error between the calculated intensities and the real intensities are superior to a certain threshold, the pixel is determined to be in the region of a shadow. The algorithm runs in loop to identify a valid set of images, that is to say without shadow, for each pixel. The method is designed to limit the computational cost. The results show that the accuracy is maintained as compared to much heavier computational cost algorithms.