Shading derivation from an unspecified object for augmented reality

We propose a method to derive shading by referring to an unspecified object to synthesize CG objects realistically into an actual scene. Since our method does not require specific light probes and can be implemented with a commercial RGB + depth sensor, it is applicable to consumer environments. The method conducts spherical harmonic (SH) basis functions regression against luminance of the reference object on the basis of its surface normal. The regression is done along with regularization that constrains the ratios of weights of SH basis functions. The regularization we force is shown to be effective in eliminating ringing artifacts of derived shading. Our method should also be robust to the changing of reference objects because it may refer to various objects. Hence, we evaluated our derived shading with a perceptual indication and showed our method is able to robustly derive perceptually diffuse shading regardless of the reference object's specular reflectance property.