Recovering Shape and Spatially-Varying Surface Reflectance under Unknown Illumination

摘要

We present a novel integrated approach for estimating bothrnspatially-varying surface reflectance and detailed geometry from arnvideo of a rotating object under unknown static illumination. Key tornour method is the decoupling of the recovery of normal and surfacernreflectance from the estimation of surface geometry. We define anrnapparent normal field with corresponding reflectance for each pointrn(including those not on the object’s surface) that best explain the observations.rnWe observe that the object’s surface goes through pointsrnwhere the apparent normal field and corresponding reflectance exhibitrna high degree of consistency with the observations. However,rnestimating the apparent normal field requires knowledge of the unknownrnincident lighting. We therefore formulate the recovery ofrnshape, surface reflectance, and incident lighting, as an iterative processrnthat alternates between estimating shape and lighting, and simultaneouslyrnrecovers surface reflectance at each step. To recoverrnthe shape, we first form an initial surface that passes through locationsrnwith consistent apparent temporal traces, followed by a refinementrnthat maximizes the consistency of the surface normals with thernunderlying apparent normal field. To recover the lighting, we relyrnon appearance-from-motion using the recovered geometry from thernprevious step. We demonstrate our integrated framework on a varietyrnof synthetic and real test cases exhibiting a wide variety ofrnmaterials and shape.