Photometric stereo with auto-radiometric calibration

We propose a novel method for estimating surface normals and a radiometric response function of a camera at the same time. Photometric stereo assumes that a camera is radiometrically calibrated in advance, so that image irradiance values can be determined from observed pixel values. Our proposed method avoids such often cumbersome radiometric calibration of a camera by the simultaneous estimation of surface normals and a radiometric response function. The key idea of our method is to make use of the consistency; the irradiance values converted from pixel values by using the radiometric response function should be equal to the corresponding irradiance values calculated from the surface normals. We show that the simultaneous estimation results in a linear least-square problem with linear constraints. The experimental results demonstrate that our method can estimate surface normals accurately even when images are captured by using cameras with nonlinear radiometric response functions.