Uncalibrated Near-Light Photometric Stereo

摘要: In this work we solve the uncalibrated photometric stereo problem with lights placed near scene. We investigate different image formation models and find one that best fits our observations. Although devised model is more complex than its far-light counterpart, show under a global linear ambiguity reconstruction possible up to rotation scaling, which can be easily fixed. also propose solution for reconstructing normal map, albedo, light positions intensities of scene given only sequence near-light images. This done in an alternating minimization framework first estimates both normals then intensities. validate method on real world experiments leads significant improvement surface compared classic distant illumination case.