Resolving the Generalized Bas-Relief Ambiguity by Entropy Minimization

Alldrin, Neil; Mallick, Satya P.; Kriegman, David

doi:10.1109/cvpr.2007.383208

Cited by 106 publications

(91 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…We are not the first to explore global parsimony priors on reflectance: different forms of this idea have been used in intrinsic images techniques [15], photometric stereo [46], shadow removal [47], and color representation [48]. We use the quadratic entropy formulation of [49] to minimize the entropy of log-reflectance, thereby encouraging parsimony.…”

Section: Parsimonymentioning

confidence: 99%

Shape, Illumination, and Reflectance from Shading

Barron¹,

Malik²

2013

197

420

View full text Add to dashboard Cite

Abstract-A fundamental problem in computer vision is that of inferring the intrinsic, 3D structure of the world from flat, 2D images of that world. Traditional methods for recovering scene properties such as shape, reflectance, or illumination rely on multiple observations of the same scene to overconstrain the problem. Recovering these same properties from a single image seems almost impossible in comparison -there are an infinite number of shapes, paint, and lights that exactly reproduce a single image. However, certain explanations are more likely than others: surfaces tend to be smooth, paint tends to be uniform, and illumination tends to be natural. We therefore pose this problem as one of statistical inference, and define an optimization problem that searches for the most likely explanation of a single image. Our technique can be viewed as a superset of several classic computer vision problems (shape-from-shading, intrinsic images, color constancy, illumination estimation, etc) and outperforms all previous solutions to those constituent problems.

show abstract

Section: Parsimonymentioning

confidence: 99%

Shape, Illumination, and Reflectance from Shading

Barron¹,

Malik²

2013

197

420

View full text Add to dashboard Cite

show abstract

“…While the constant albedo assumption is often plausible in lunar and planetary mapping (Horn, 1990;Kirk et al, 2003), previous works on natural image statistics (Huang and Mumford, 1999) and Shape from Shading (Alldrin et al, 2007;Barron and Malik, 2011) assume a specific statistical behaviour of the albedo, that albedo values tends to be highly concentrated and the difference between adjacent albedo is very small, effectively suggesting a piece-wise constant albedo assumption.…”

Section: Locally Varying Albedomentioning

confidence: 99%

SHAPE AND ALBEDO FROM SHADING (SAfS) FOR PIXEL-LEVEL DEM GENERATION FROM MONOCULAR IMAGES CONSTRAINED BY LOW-RESOLUTION DEM

Wu¹,

Liu²,

Grumpe³

et al. 2016

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

ABSTRACT:Lunar topographic information, e.g., lunar DEM (Digital Elevation Model), is very important for lunar exploration missions and scientific research. Lunar DEMs are typically generated from photogrammetric image processing or laser altimetry, of which photogrammetric methods require multiple stereo images of an area. DEMs generated from these methods are usually achieved by various interpolation techniques, leading to interpolation artifacts in the resulting DEM. On the other hand, photometric shape reconstruction, e.g., SfS (Shape from Shading), extensively studied in the field of Computer Vision has been introduced to pixellevel resolution DEM refinement. SfS methods have the ability to reconstruct pixel-wise terrain details that explain a given image of the terrain. If the terrain and its corresponding pixel-wise albedo were to be estimated simultaneously, this is a SAfS (Shape and Albedo from Shading) problem and it will be under-determined without additional information. Previous works show strong statistical regularities in albedo of natural objects, and this is even more logically valid in the case of lunar surface due to its lower surface albedo complexity than the Earth. In this paper we suggest a method that refines a lower-resolution DEM to pixel-level resolution given a monocular image of the coverage with known light source, at the same time we also estimate the corresponding pixel-wise albedo map. We regulate the behaviour of albedo and shape such that the optimized terrain and albedo are the likely solutions that explain the corresponding image. The parameters in the approach are optimized through a kernel-based relaxation framework to gain computational advantages. In this research we experimentally employ the Lunar-Lambertian model for reflectance modelling; the framework of the algorithm is expected to be independent of a specific reflectance model. Experiments are carried out using the monocular images from Lunar Reconnaissance Orbiter (LRO) Narrow Angle Camera (NAC) (0.5 m spatial resolution), constrained by the SELENE and LRO Elevation Model (SLDEM 2015) of 60 m spatial resolution. The results indicate that local details are largely recovered by the algorithm while low frequency topographic consistency is affected by the low-resolution DEM.

show abstract

“…To our knowledge, this very general radiance model has not been used in the context of photometric stereo so far. In most applications, a single point light source at infinity is assumed, so that each lighting is represented by a constant vector in R 3 [17,7,1,14,15]. Closer to this general expression is the important work of Basri et al in [2], where a distribution of point light sources at infinity is assumed, allowing the authors to develop shape recovery algorithms based on low order spherical harmonic decompositions.…”

Section: Image Formation Modelmentioning

confidence: 99%

“…As suggested in literature [1,15], we restrict our study to the case of piecewisesmooth ("Arlequin-like") albedos, which can be enforced by a TV semi-norm as well:…”

Section: Continuous Variational Energymentioning

confidence: 99%

A $$L^1$$-TV Algorithm for Robust Perspective Photometric Stereo with Spatially-Varying Lightings

Quéau

Lauze

Durou

2015

Lecture Notes in Computer Science

View full text Add to dashboard Cite

OATAO is an open access repository that collects the work of Toulouse researchers and makes it freely available over the web where possible. This is an author-deposited version published in : http://oatao.univ-toulouse.fr/ Eprints ID : 15234The contribution was presented at SSVM 2015 :http://ssvm2015.math.u-bordeaux.fr/ Abstract. We tackle the problem of perspective 3D-reconstruction of Lambertian surfaces through photometric stereo, in the presence of outliers to Lambert's law, depth discontinuities, and unknown spatiallyvarying lightings. To this purpose, we introduce a robust L 1 -TV variational formulation of the recovery problem where the shape itself is the main unknown, which naturally enforces integrability and permits to avoid integrating the normal field.

show abstract

Resolving the Generalized Bas-Relief Ambiguity by Entropy Minimization

Cited by 106 publications

References 14 publications

Shape, Illumination, and Reflectance from Shading

Shape, Illumination, and Reflectance from Shading

SHAPE AND ALBEDO FROM SHADING (SAfS) FOR PIXEL-LEVEL DEM GENERATION FROM MONOCULAR IMAGES CONSTRAINED BY LOW-RESOLUTION DEM

A $$L^1$$-TV Algorithm for Robust Perspective Photometric Stereo with Spatially-Varying Lightings

Contact Info

Product

Resources

About