Abstract
Uncalibrated photometric stereo aims at estimating the 3D-shape of a surface, given a set of images captured from the same viewing angle, but under unknown, varying illumination. While the theoretical foundations of this inverse problem under directional lighting are well-established, there is a lack of mathematical evidence for the uniqueness of a solution under general lighting. On the other hand, stable and accurate heuristical solutions of uncalibrated photometric stereo under such general lighting have recently been proposed. The quality of the results demonstrated therein tends to indicate that the problem may actually be well-posed, but this still has to be established. The present paper addresses this theoretical issue, considering first-order spherical harmonics approximation of general lighting. Two important theoretical results are established. First, the orthographic integrability constraint ensures uniqueness of a solution up to a global concave–convex ambiguity, which had already been conjectured, yet not proven. Second, the perspective integrability constraint makes the problem well-posed, which generalizes a previous result limited to directional lighting. Eventually, a closed-form expression for the unique least-squares solution of the problem under perspective projection is provided, allowing numerical simulations on synthetic data to empirically validate our findings.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Joyful Yell: https://www.thingiverse.com/thing:897412; other datasets: http://www-graphics.stanford.edu/data/3dscanrep.
References
Bartal O, Ofir N, Lipman Y, Basri R (2018) Photometric stereo by hemispherical metric embedding. J Math Imaging Vis 60(2):148–162
Basri R, Jacobs DW (2003) Lambertian reflectances and linear subspaces. IEEE Trans Pattern Anal Mach Intell 25(2):218–233
Basri R, Jacobs DW, Kemelmacher I (2007) Photometric stereo with general, unknown lighting. Int J Comput Vis 72(3):239–257
Belhumeur PN, Kriegman DJ, Yuille AL (1999) The bas-relief ambiguity. Int J Comput Vis 35(1):33–44
Breuß M, Cristiani E, Durou J-D, Falcone M, Vogel O (2012) Perspective shape from shading: ambiguity analysis and numerical approximations. SIAM J Imaging Sci 5(1):311–342
Chen G, Han K, Shi B, Matsushita Y, Wong K-YK (2019) Self-calibrating deep photometric stereo networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), pp 8739–8747
Chen L, Zheng Y, Shi B, Subpa-Asa A, Sato I (2019) A microfacet-based model for photometric stereo with general isotropic reflectance. IEEE Trans Pattern Anal Mach Intell (in press)
Durou J-D, Falcone M, Sagona M (2008) Numerical methods for shape-from-shading: a new survey with benchmarks. Comput Vis Image Underst 109(1):22–43
Einstein A (1905) Zur elektrodynamik bewegter körper. Annalen der physik 322(10):891–921
Frolova D, Simakov D, Basri R (2004) Accuracy of spherical harmonic approximations for images of Lambertian objects under far and near lighting. In: Proceedings of the European conference on computer vision (ECCV), pp 574–587
Haefner B, Ye Z, Gao M, Wu T, Quéau Y, Cremers D (2019) Variational uncalibrated photometric stereo under general lighting. In: Proceedings of the international conference on computer vision (ICCV)
Hayakawa H (1994) Photometric stereo under a light source with arbitrary motion. J Opt Soc Am A 11(11):3079–3089
Ikehata S (2018) CNN-PS: CNN-based photometric stereo for general non-convex surfaces. In: Proceedings of the European conference on computer vision (ECCV), pp 3–18
Jaffe A (2013) Lorentz transformations, rotations, and boosts, 2013. Online notes. http://home.ku.edu.tr/~amostafazadeh/phys517_518/phys517_2016f/Handouts/A_Jaffi_Lorentz_Group.pdf. Accessed Sept 2019
Khanian M, Boroujerdi AS, Breuß M (2018) Photometric stereo for strong specular highlights. Comput Visual Media 4(1):83–102
Kozera R, Prokopenya A (2018) Second-order algebraic surfaces and two image photometric stereo. In: International conference on computer vision and graphics, pp 234–247
Li J, Robles-Kelly A, You S, Matsushita Y (2019) Learning to minify photometric stereo. In: Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), pp 7568–7576
Logothetis F, Mecca R, Quéau Y, Cipolla R (2016) Near-field photometric stereo in ambient light. In: Proceedings of the British machine vision conference (BMVC)
Longuet-Higgins HC (1987) A computer algorithm for reconstructing a scene from two projections. In: Fischler MA, Firschein O (eds) Readings in computer vision. Morgan Kaufmann, Burlington, pp 61–62
Mecca R, Falcone M (2013) Uniqueness and approximation of a photometric shape-from-shading model. SIAM J Imaging Sci 6(1):616–659
Mecca R, Wetzler A, Bruckstein AM, Kimmel R (2014) Near field photometric stereo with point light sources. SIAM J Imaging Sci 7(4):2732–2770
Mecca R, Quéau Y, Logothetis F, Cipolla R (2016) A single-lobe photometric stereo approach for heterogeneous material. SIAM J Imaging Sci 9(4):1858–1888
Mo Z, Shi B, Lu F, Yeung S-K, Matsushita Y (2018) Uncalibrated photometric stereo under natural illumination. In: Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), pp 2936–2945
Onn R, Bruckstein A (1990) Integrability disambiguates surface recovery in two-image photometric stereo. Int J Comput Vis 5(1):105–113
Papadhimitri T, Favaro P (2013) A new perspective on uncalibrated photometric stereo. In: Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), pp 1474–1481
Papadhimitri T, Favaro P (2014) Uncalibrated near-light photometric stereo. In: Proceedings of the British machine vision conference (BMVC)
Peng S, Haefner B, Quéau Y, Cremers D (2017) Depth super-resolution meets uncalibrated photometric stereo. In: Proceedings of the IEEE international conference on computer vision (ICCV) workshops, pp 2961–2968
Poincaré H (1905) Sur la dynamique de l’électron. Comptes Rendus de l’Académie des Sciences 140:1504–1508
Quéau Y, Mecca R, Durou J-D, Descombes X (2017) Photometric stereo with only two images: a theoretical study and numerical resolution. Image Vis Comput 57:175–191
Quéau Y, Wu T, Cremers D (2017) Semi-calibrated near-light photometric stereo. In: International conference on scale space and variational methods in computer vision (SSVM), pp 656–668
Quéau Y, Wu T, Lauze F, Durou J-D, Cremers D (2017) A non-convex variational approach to photometric stereo under inaccurate lighting. In: Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), pp 99–108
Quéau Y, Durix B, Wu T, Cremers D, Lauze F, Durou J-D (2018) LED-based photometric stereo: modeling, calibration and numerical solution. J Math Imaging Vis 60(3):313–340
Quéau Y, Durou J-D, Aujol J-F (2018) Normal integration: a survey. J Math Imaging Vis 60(4):576–593
Radow G, Hoeltgen L, Quéau Y, Breuß M (2019) Optimisation of classic photometric stereo by non-convex variational minimisation. J Math Imaging Vis 61(1):84–105
Ramamoorthi R, Hanrahan P (2001) An efficient representation for irradiance environment maps. In: Proceedings of the annual conference on computer graphics and interactive techniques, pp 497–500
Sengupta S, Zhou H, Forkel W, Basri R, Goldstein T, Jacobs D (2018) Solving uncalibrated photometric stereo using fewer images by jointly optimizing low-rank matrix completion and integrability. J Math Imaging Vis 60(4):563–575
Shi B, Inose K, Matsushita Y, Tan P, Yeung S-K, Ikeuchi K (2014) Photometric stereo using internet images. In: Proceedings of the international conference on 3D vision (3DV), vol 1, pp 361–368
Shi B, Wu Z, Mo Z, Duan D, Yeung S-K, Tan P (2019) A benchmark dataset and evaluation for non-Lambertian and uncalibrated photometric stereo. IEEE Trans Pattern Anal Mach Intell 41(2):271–284
Tozza S, Falcone M (2016) Analysis and approximation of some shape-from-shading models for non-Lambertian surfaces. J Math Imaging Vis 55(2):153–178
Tozza S, Mecca R, Duocastella M, Del Bue A (2016) Direct differential photometric stereo shape recovery of diffuse and specular surfaces. J Math Imaging Vis 56(1):57–76
Triggs B, McLauchlan PF, Hartley RI, Fitzgibbon AW (1999) Bundle adjustment – a modern synthesis. In: International workshop on vision algorithms (ICCV workshops), pp 298–372
Woodham RJ (1978) Reflectance map techniques for analyzing surface defects in metal castings. Technical report AITR-457. MIT
Woodham RJ (1980) Photometric method for determining surface orientation from multiple images. Opt Eng 19(1):139–144
Yuille A, Snow D (1997) Shape and albedo from multiple images using integrability. In: Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), pp 158–164
Zhang R, Tsai P-S, Cryer JE, Shah M (1999) Shape-from-shading: a survey. IEEE Trans Pattern Anal Mach Intell 21(8):690–706
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Appendices
Appendix 1: Proof of Proposition 5.1
Proposition 5.1 characterizes the integrability of a normal field in terms of the coefficients \(m_1\), \(m_2\) and \(m_3\) of \(\mathbf {m}:=\rho \mathbf {n}\). The following proof of this proposition is largely inspired by [25].
Proof
According to Eqs. (5.15)–(5.17), integrability of the normal field under perspective projection can be written as:
where \(\times \) denotes the cross-product.
Besides, \(-\dfrac{\mathbf {m}}{m_3} = \begin{pmatrix} p \\ q \\ -1 \end{pmatrix}\) according to (5.14). If we denote \(\mathbf {w}_1 = \left[ 0,-f,v\right] ^\top \) and \(\mathbf {w}_2 = \left[ -f,0,u\right] ^\top \), then (5.75) yields the following equation over \(\Omega \):
Multiplying Eq. (5.76) by \({m_3}^3\):
In addition, \((\mathbf {w}_1 \times \mathbf {m}) \perp \mathbf {m}\) and \((\mathbf {w}_2 \times \mathbf {m}) \perp \mathbf {m}\), thus the following relationship holds over \(\Omega \):
which concludes the proof. \(\square \)
Appendix 2: Proof of Theorem 5.1
Theorem 5.1 characterizes scaled Lorentz transformations. Its proof relies on the following Propositions 5.2–5.4 from Lorentz’ group theory (proofs of these propositions can be found in [14]).
Proposition 5.2
For any proper and orthochronous Lorentz transformation \(\mathbf {A} \in L^{p}_{o}\), there exists a unique couple \((\mathbf {v}, \mathbf {O}) \in B(\mathbf {0},1) \times SO(3,\mathbb {R})\) such that
where \(\gamma = \frac{1}{\sqrt{1-\left\Vert \mathbf {v}\right\Vert ^2}}\), and
Proposition 5.3
The product of two proper/improper transformations is a proper one, and the product of a proper and improper transformation is an improper one. The same for the orthochronous property.
Proposition 5.4
Matrix \(\mathbf {T} = \begin{pmatrix} -1 &{} 0 &{} 0 &{} 0\\ 0 &{} 1 &{} 0 &{} 0 \\ 0 &{} 0 &{} 1 &{} 0 \\ 0 &{} 0 &{} 0 &{} 1 \end{pmatrix}\) is improper and non-orthochronous, and
matrix \(\mathbf {P} = \begin{pmatrix} 1 &{} 0 &{} 0 &{} 0\\ 0 &{} -1 &{} 0 &{} 0 \\ 0 &{} 0 &{} -1 &{} 0 \\ 0 &{} 0 &{} 0 &{} -1 \end{pmatrix}\) is improper and orthochronous.
Using these already known results, we propose the following characterization of Lorentz transformations:
Proposition 5.5
For any Lorentz transformation \(\mathbf {A} \in L\), there exists a unique couple \((\mathbf {v}, \mathbf {O}) \in B(\mathbf {0},1) \times SO(3,\mathbb {R})\) such that
Proof
We first assume that \(\mathbf {A} \in L^{i}_{n}\).
According to Proposition 5.4, \(\mathbf {T} \in L^{i}_{n}\). Thus using Proposition 5.3: \(\mathbf {T}\mathbf {A} \in L^{p}_{o}\). Therefore, according to Proposition 5.2, there exists a unique couple \((\mathbf {v}, \mathbf {O}) \in B(\mathbf {0},1) \times SO(3,\mathbb {R})\) such that \( \mathbf {T}\mathbf {A} = \mathbf {S}(\mathbf {v})\mathbf {R}(\mathbf {O})\). Since \(\mathbf {T}\mathbf {T} = \mathbf {I_4}\), this implies that \(\mathbf {A} = \mathbf {T}\mathbf {S}(\mathbf {v})\mathbf {R}(\mathbf {O})\). In addition, \(\epsilon _1(\mathbf {A}) = -1\) and \(\epsilon _2(\mathbf {A}) = 1\), hence:
With the same reasoning, we get the result for all the other transformations. \(\square \)
Combining Proposition 5.5 and the definition (5.24) of scaled Lorentz transformations, we get Theorem 5.1.
Appendix 3: Some Useful Results on GBR and Lorentz Matrices, and Corollary 5.1
The aim of this section is to prove Corollary 5.1, which was used in the proofs of Theorems 5.2 and 5.3. Its proof relies on a few results on GBR and Lorentz matrices, which we provide in the following.
Let us denote by G the group of GBR transformations, and by \(G_s\) that of scaled GBR transformations defined by
Both are subgroups of \(GL(3,\mathbb {R})\) under the matrix product. For all \(\mathbf {B} = s\mathbf {A} \in G_s\), we call s the scale part of \(\mathbf {B}\), and \(\mathbf {A}\) its GBR part.
Let \(\mathbf {C} \in \mathbb {R}^{n \times n}\) with \(n > 1\) and \(C_{ij}\) its entries. We will use the following notation for a minor of size two:
where \(1 \le i < k \le n\) and \(1 \le j < l \le n\).
Such minors allow to characterize scaled GBR matrices:
Proposition 5.6
Let \(\mathbf {A} \in \mathbb {R}^{3 \times 3}, A_{ij}\) the entries of \(\mathbf {A}\). Then, \(\mathbf {A}\) is a scaled GBR transformation iff \(\mathbf {A}\) is invertible and fulfills the following equations:
Proof
See [4]. \(\square \)
Proposition 5.7
Let \(\mathbf {v} \in B_{0}(1), \gamma = \frac{1}{\sqrt{1-\left\Vert \mathbf {v}\right\Vert ^2}}\), then \(\mathbf {C} = \mathbf {I_3} + \frac{\gamma ^{2}}{1 + \gamma } \mathbf {v} \mathbf {v}^\top \) is positive definite.
Proof
Let \(\mathbf {B} = \frac{\gamma ^{2}}{1 + \gamma } \mathbf {v} \mathbf {v}^\top \). We note \(E_\lambda (\mathbf {B})\) the eigenspace associated to the eigenvalue \(\lambda \) of \(\mathbf {B}\). \(\mathbf {B}\) is symmetric, thus according to the spectral theorem, all the eigenvalues of \(\mathbf {B}\) are real, and \(\mathbb {R}^3 = \bigoplus \limits _{i=1}^r E_{\lambda _i}(\mathbf {B})\) with \(r \le 3\) the number of eigenvalues, and \(\lbrace \lambda _i \rbrace _{i=1 \ldots r}\) the eigenvalues of \(\mathbf {B}\). Hence: \(\text {dim}(\mathbb {R}^3) = \sum \limits _{i=1}^r \text {dim}(E_{\lambda _i}(\mathbf {B}))\). According to the rank-nullity theorem, \(\text {dim}(Ker(\mathbf {B})) + rank(\mathbf {B}) = 3\), and by definition \(rank(\mathbf {B}) = 1\), thus \(\text {dim}(Ker(\mathbf {B})) = \text {dim}(E_0(\mathbf {B})) = 2\). We deduce that there exists a unique nonzero eigenvalue \(\lambda \in \mathbb {R}\backslash \{0\}\) such that \(\mathbb {R}^3 = E_0(\mathbf {B}) \bigoplus E_\lambda (\mathbf {B}) \) with \(\text {dim}(E_\lambda (\mathbf {B})) = 1\).
Let \(\Pi _{\mathbf {v}}\) the orthogonal projection onto \(span \lbrace \mathbf {v} \rbrace \), and let \(\mathbf {x} \in \mathbb {R}^3\) : \(\Pi _{\mathbf {v}}(\mathbf {x}) = \frac{\mathbf {v}\mathbf {v}^\top }{\left\Vert \mathbf {v}\right\Vert ^2} \mathbf {x}\) and \(\Pi _{\mathbf {v}}(\mathbf {v}) = \mathbf {v}\). We have: \(\mathbf {B} = \frac{\gamma ^{2}}{1 + \gamma }\left\Vert \mathbf {v}\right\Vert ^2\Pi _{\mathbf {v}}\) and \(\mathbf {B}\mathbf {v} = (\frac{\gamma ^{2}}{1 + \gamma }\left\Vert \mathbf {v}\right\Vert ^2)\mathbf {v} \). Thus, \(\frac{\gamma ^{2}}{1 + \gamma }\left\Vert \mathbf {v}\right\Vert ^2\) is an eigenvalue of \(\mathbf {B}\) and \(\lambda = \frac{\gamma ^{2}}{1 + \gamma }\left\Vert \mathbf {v}\right\Vert ^2\) . Besides, \(\frac{\lambda }{\gamma -1} = \frac{\gamma ^{2}}{(\gamma - 1)(\gamma + 1)}\left\Vert \mathbf {v}\right\Vert ^2 = \frac{\gamma ^{2}}{\gamma ^2 - 1}\left\Vert \mathbf {v}\right\Vert ^2 = \frac{1}{1 - \frac{1}{\gamma ^2}}\left\Vert \mathbf {v}\right\Vert ^2 = \frac{1}{1-(1-\left\Vert \mathbf {v}\right\Vert ^2)}\left\Vert \mathbf {v}\right\Vert ^2=1\). Therefore, \(\lambda = \gamma - 1 \) and the eigenvalues of \(\mathbf {B}\) are 0 and \((\gamma - 1)\). Let \(\alpha \in \lbrace 0,\gamma - 1 \rbrace \) and \(\mathbf {u} \in E_\alpha (\mathbf {B})\). We have:
Thus, \(1>0\) and \(\gamma > 0\) are the eigenvalues of \(\mathbf {C}\) with \(E_1(\mathbf {C}) = E_0(\mathbf {B})\) and \(E_\gamma (\mathbf {C}) = E_{\gamma -1}(\mathbf {B})\). Consequently, \(\mathbf {C}\) is positive definite. \(\square \)
Proposition 5.8
Let \(\mathbf {A}_s \in L_s\) a scaled Lorentz transformation. The submatrix \(\mathbf {B}\) formed by the last 3 rows and 3 columns of \(\mathbf {A}_s\) is invertible.
Proof
By definition of \(\mathbf {A}_s\), there exists a unique couple \((s,{\tilde{\mathbf {A}}}) \in \mathbb {R}\backslash \{0\} \times L\) such that \(\mathbf {A}_s = s{\tilde{\mathbf {A}}}\). Hence, from Proposition 5.5, there exists a unique couple \((\mathbf {v}, \mathbf {O}) \in B(\mathbf {0},1) \times SO(3,\mathbb {R})\) such that \( \mathbf {B} = s \epsilon _{2}({\tilde{\mathbf {A}}})(\mathbf {I_{3}} + \frac{\gamma ^{2}}{1 + \gamma } \mathbf {v} \mathbf {v}^\top ) \mathbf {O}\). Since \(\mathbf {O} \in SO(3,\mathbb {R})\), \(\text {det}(\mathbf {O}) = 1\). In addition, Proposition 5.7 implies \(\text {det}(\mathbf {I_{3}} + \frac{\gamma ^{2}}{1 + \gamma } \mathbf {v} \mathbf {v}^\top ) {>} 0\), thus \(\text {det}(\mathbf {B}) \ne 0\). \(\square \)
Corollary 5.1
Let \(\mathbf {A}_s \in L_s\) a scaled Lorentz transformation. If its entries \(A_{ij}\) fulfill
then the submatrix \(\mathbf {B}\) of \(\mathbf {A}_s\) formed by the last 3 rows and 3 columns is a scaled GBR, i.e., there exists a unique quadruple \((\lambda ,\mu ,\nu ,\beta ) \in \mathbb {R}^4\) with \(\lambda \ne 0, \beta \ne 0\) such that
Proof
According to Proposition 5.8, \(\mathbf {B}\) is invertible. Besides, \(\mathbf {A}_s\) fulfill Eqs. (5.87) iff \(\mathbf {B}\) fulfill Eqs. (5.85), thus according to Proposition 5.6, \(\mathbf {B} \in G_s\). \(\square \)
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this chapter
Cite this chapter
Brahimi, M., Quéau, Y., Haefner, B., Cremers, D. (2020). On the Well-Posedness of Uncalibrated Photometric Stereo Under General Lighting. In: Durou, JD., Falcone, M., Quéau, Y., Tozza, S. (eds) Advances in Photometric 3D-Reconstruction. Advances in Computer Vision and Pattern Recognition. Springer, Cham. https://doi.org/10.1007/978-3-030-51866-0_5
Download citation
DOI: https://doi.org/10.1007/978-3-030-51866-0_5
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-51865-3
Online ISBN: 978-3-030-51866-0
eBook Packages: Computer ScienceComputer Science (R0)