Professional Documents
Culture Documents
com
Alexander Prokhorov
Virial International, LLC, 538 Palmspring Drive, Gaithersburg, Maryland 20878,
USA (avpro@virial.com)
This paper proposes a three-component bidirectional reflectance distribution function (3C BRDF) model
consisting of diffuse, quasi-specular, and glossy components for calculation of effective emissivities of
blackbody cavities and then investigates the properties of the new reflection model. The particle swarm
optimization method is applied for fitting a 3C BRDF model to measured BRDFs. The model is incorpo-
rated into the Monte Carlo ray-tracing algorithm for isothermal cavities. Finally, the paper compares the
results obtained using the 3C model and the conventional specular-diffuse model of reflection. © 2012
Optical Society of America
OCIS codes: 120.5630, 230.6080, 290.1483.
DHR
the USD model. The BRDF of the diffuse (Lamber-
0.88
tian) component is the constant:
0.86 0.001
0.005
0.84 0.01
R
d; 0.82 0.05
f r;d (7)
π
0.1
0.8 0.2
0.78
0 10 20 30 40 50 60 70 80 90
where Rd is the diffuse reflectance (DHR) of the
Incidence Angle (°)
diffuse component.
For quasi-specular and glossy components, the Fig. 2. (Color online) Dependences of the DHR on the inci-
dence angle computed for the Geisler-Moroder and Dür BRDF
model proposed by Geisler-Moroder and Dür [28]
model for R 1 and σ 0.0001, 0.005, 0.01, 0.05, 0.1, and 0.2
for rough surfaces has been chosen in its isotropic (see legend).
version:
Rθh tan θh 2
f r θi ; θv ; ϕ exp −
πσ σ where Rqs is the specular reflectance of a microfacet
21 cos θi cos θv − sin θi sin θv cos ϕ at the normal incidence.
× ; We also preferred Schlick’s approximation over
cos θi cos θv 4
Fresnel’s equation for the following reasons.
(8)
1. In the fitting problem, it is important to know a
where σ is the roughness parameter and θh is the good initial value of a fitted parameter or at least the
halfway angle (Fig. 1). narrowest range of its acceptability. These data are
This BRDF model is based on the microfacet model unavailable a priori for the refractive index n and
of reflection [29]. The last multiplier allows us to the extinction coefficient k while for the reflectance
avoid infinite BRDF values at θi θv π∕2 and, at the normal incidence, 0 ≤ Rqs ≤ 1.
at the same time, plays the role of normalization fac- 2. Fresnel’s equation has more complicated form
tor. Specular reflectance, R, of microfacet depends on and comprises n2 and k2 . These factors result in an
the angle θh of the incidence onto the microfacet. This increase of number of objective function minima and
angle, called the halfway angle, is formed by the sur- require introducing additional criteria to select the
face normal and the halfway vector h 2‖ω ωv −ωi global minimum.
v −ωi ‖ 3. Although Eq. (8) is based on the microfacet
(Fig. 1). It is assumed that microfacet normals have model for rough surfaces, we found that one can
Gaussian distribution with zero mean value and apply this equation for modeling of reflection with a
standard deviation σ. The greater the σ the greater different scattering nature (e.g., subsurface or volu-
the surface roughness and, correspondingly, the metric scattering for quasi-specular black paints).
wider the BRDF lobe. Elementary physical reason- In such cases, parameters n and k can take uncommon
ing suggests that DHR of a microfacet model must values, lose their physical meaning, and should be
equal 1 for all incidence angles if Rθh ≡ 1. DHRs, considered as the fitting parameters only. Parameter
of all microfacet models that neglect multiple Rqs remains within interval 0; 1 and maintains
reflections among microfacets, deviate from 1 for its physical meaning. Indeed, in such a case, σ in
such a case; the greater roughness, the greater this Eq. (8) should be considered only as a parameter
deviation. determining the width of the BRDF lobe.
The last term in Eq. (8) is constructed in such a 4. The Schlick formula not only approximates
way that the deviation of DHR from 1 at Rθh ≡ 1 Fresnel’s law accurately enough for dielectrics but
is minimized (Fig. 2). Deviations become significant also works several times faster.
only for large σ and large θi . Since the reflectance of
materials of blackbody cavities is essentially low, de- For nondielectrics, Eq. (9) gives deviations from
viations of DHR computed by integration of BRDF in Fresnelian reflectance at large θh . However, large
Eq. (8) from the actual DHR values will be substan- θh values usually correspond to very rough surfaces
tially less. for which the effect of multiple reflections among mi-
The original model [28]) uses Schlick’s approxima- crofacets overrides Schlick’s approximation errors.
tion [30] of Fresnel’s reflection law for unpolarized The numerical experiments we performed did not re-
light: veal significant differences between the employment
of Fresnel’s law and Schlick’s approximation for the
tasks considered in this paper. The only drawback
Rθh Rqs 1 − Rqs 1 − cos θh 5 ; (9) with Schlick’s approximation, which hampers the
40
BRDF (sr -1)
30
20
10
0
-90 -60 -30 0 30 60 90
Viewing Angle (°)
Fig. 3. (Color online) In-plane Cartesian plots for the Geisler-
Moroder and Dür BRDF model with R 0.1, σ 0.1, and six in-
cidence angles shown in legend. The negative values of viewing
angles correspond to backscattering.
where every summand is the DHR of the appropriate The double integrals in Eqs. (12) and (13)
component: were computed numerically using an adaptive
sin2 θv uθ ; (20)
-75 -60 -45 -30 -15 0 15 30 45 60 75
Viewing Angle (°)
ϕv 2πuϕ ; (21)
Fig. 5. (Color online) Results for the 3C BRDF model (lines) fitted
to the Chemglaze Z302 black paint BRDFs measured at 10.6 μm
(symbols). where uθ and uϕ are two random numbers from a uni-
formly distributed set between 0 and 1, θv and ϕv are
the polar and azimuthal angles of a reflected ray in
the local spherical coordinate system with the origin
emissivity that imitates viewing conditions with the
at the point of reflection and z axis coincident with
objective lens or other imaging optical system, can be
the surface normal.
modeled by generating rays uniformly distributed
We used an alternative method that produces Car-
within the conical solid angle and starting from
tesian coordinates of the unit vector directed along
the focal point (cone apex). If the starting point the reflected ray. The pair of uniformly distributed
and the directional vector are known, one must solve random numbers ux and uy are generated and ac-
the equation for a cavity surface together with para- cepted if u2x u2y < 1, otherwise the new pair of ux
metric equations for a straight line. If the cavity con- and uy are generated. Cartesian coordinates of the
sists of many surfaces, the line-surface intersection unit reflection vector ωv in the local coordinate sys-
points have to be found for each surface. The true tem are computed as
solution corresponds to the point that lies on some
surface within its bounds and has the minimal dis- 8
tance from the starting point. We implemented this < ωx 2ux − 1
>
ωy 2uy − 1 22
algorithm for cavities formed by the revolving of q .
>
: ω 1 − ω2 − ω2
the non-self-intersecting polygonal line around the z x y
cavity axis.
The statistical weight w Rd is assigned to the
diffusely reflected ray.
0.3 The sampling procedure described in [28] was ap-
DHR (total) plied to quasi-specular and glossy components. First,
0.25
DHR (diffuse) spherical coordinates of the halfway vector h (Fig. 1)
DHR (quasi-specular) are computed using random numbers uθ and uϕ :
DHR (glossy)
0.2 θh tan−1 −σ ln uθ (23)
and
DHR
0.15
ϕh 2πuϕ : (24)
0.1
(ω
vx ωix − 2ωi · hhx
ωvy ωiy − 2ωi · hhy . 26 l
ωvz ωiz − 2ωi · hhz
B
According to the sampling procedure proposed in rc ra
[31], the statistical weight
A
2Rθh
w 27
1 − ωiz ∕ωvz l
1X n Y
m k
0.99998
0.99996 0.999
0.005
0.998
0.99992 0.007
0.01
0.99990
0.997
0.99988
0.2
0.99986 0.3
0.996
0.4
0.99984 0.5
Diffuse
0.99982 0.995
0 2 4 6 8 10 12 14 16 18 20
0.99980 A (°)
0.4
of reflection. We found that there are cases when
the USD model of reflection cannot reproduce results
0.3
obtained using 3C BRDF model.
The USD model remains highly valuable in assess-
0.2
ment of effective emissivities of blackbody cavities
especially at the design stage, due to the minimal
0.1
number of model’s parameters and the simplicity
of the interpretation of the results obtained. How-
0
0 10 20 30 40 50 60 70 80 90 ever, in some cases, the USD model might be too
Incidence Angle (°) crude for precise characterization of blackbody cali-
Fig. 11. (Color online) Dependences of DHR on the incidence an- bration sources. This, in turn, may affect the mea-
gle for the 3C BRDF model and its components (see legend) used in surement precision of such quantities as spectral
numerical experiments. radiance and radiance temperature.
In the future works, we plan to extend our analysis
to isothermal cavities and introduce normalized
1.0000 modified version of the RPV model instead of the
Lambertian component.
0.9998
0.9996
References
1. J. Hollandt, J. Seidel, R. Klein, G. Ulm, A. Migdall, and M.
0.9994 Ware, “Primary sources for use in radiometry,” Optical Radio-
metry, A. C. Parr, R. U. Datla, and J. L. Gardner, eds.
0.9992 (Academic, 2005), pp. 213–290.
Conical Effective Emissivity