High-frequency shape and albedo from shading using natural image statistics

摘要: We relax the long-held and problematic assumption in shape-from-shading (SFS) that albedo must be uniform or known, address problem of “shape from shading” (SAFS). Using models normally reserved for natural image statistics, we impose “naturalness” priors over shape a scene, which allows us to simultaneously recover most likely explain single image. A simplification our algorithm solves classic SFS, SAFS can solve intrinsic decomposition problem, as it superset problem. present results SAFS, on real lunar imagery Apollo missions, own pseudo-synthetic dataset, subset MIT Intrinsic Images dataset[15]. Our one unified technique appears outperform previous best individual algorithms all three tasks. coarse observation (from laser rangefinder stereo algorithm, etc) incorporated priori. demonstrate even small amount low-frequency information dramatically improves performance, motivate usage shading high-frequency (and albedo) recovery.