pdel,
e, the de-shading
oved precision of
training frame, a
s. In the working
irface reflectance
Synthetic images
nents. The results
nt information in
lation, etc.), the
it. Therefore, to
f great important.
ising images with
roblem, we need
ges (de-shading),
th an assigned
'oper photometric
precision of the
TION
ie signal recorded
es the visual light
je reflected light
1e surface. If the
at will remain?
Because no light
we consider the
uld be something
ation of a visual
f the light source,
, the geometric
odel information,
oise information
n were taken out,
ords, some of the
is possible to be
get some of the
to obtain if there
ied the process of
on as de-shading.
he inputs-outputs
itural expressive
ain the original
in the imagery.
nd its associated
iding is to obtain
real image, the
direction of the
hich has a higher
lage.
B eu cooper oc mec modems tcc i we
Inputs and outputs: The inputs of a de-shading system are:
some candidate photometric models, a brightness image and its
associated approximate height image (range image or DTM
image). The outputs are: the light source, the photometric
model approximating the brightness image, the albedo image or
a
Re rare oies rem bof nee ia ae
the surface reflectance properties image (the value of each pixel
in each image represents the albedo value or reflectance
property value at its coordinates), the improved height image
which is more precise than the approximate input height image.
light source brightness approximate
direction image height image
(y) I( x, y) Z(x, y)
1
1
i
i
i
i
i
i
i
l
i
I
i
i
1
l
1
1
zz
candidate
photometric
model 1
gradient images
p(x,y)andq(x, y)
algorithm DRP [
algorithm DPM
for determining the approximate
for determining the reflectance
properties 0 (x, y), sets Ox (x, »
photometric model
obtained approximate
shading algorithm to
obtain shaded image
photometric model
l'(x,y)
em WN Y, y) t I'(x, »)Ÿ
e decreases
9
|
algorithm DHI (SFS)
for determining updated height image Z'( x, y)
and gradient images p'(x, y), qx. y)
A
output the obtained
photometric model,
photometric parameters
$1 (x, y), ty $i x, y»
updated height image Z'( x, y),
and gradient images p' (x, y), q'(x, y)
end
WORKING FRAME
— [Bue
Fig.2. A de-shading system.
3. DE-SHADING SYSTEM
A de-shading system is proposed in the paper to perform the
de-shading task. As described in the above section, the task of
de-shading is to obtain the surface reflectance properties, the
light source, the improved higher precision height image and
the approximate photometric model from a given brightness
image and its associated approximate height image. Obviously,
this is not easy. Fortunately, there exist some methods being
more or less relevant to parts of the task. E.g., the least square
fitting method is used to calculate the surface reflectance
properties (Ikeuchi and Sato, 1991; Kay and Caelli, 1994), SFS
is used to obtain the surface orientation and height image
(Ikeuchi and Horn, 1981; Pentland, 1984; Brooks and Horn,
1985; Lee and Rosenfeld, 1985; Zheng and Cellappa, 1991;
Kimmel and Bruckstein, 1995). But all the methods need some
strong preconditions. Algorithms for calculating surface
reflectance properties assume that the photometric model is
known and that the reflectance properties are homogeneous
over the entire surface, also a precision range image is required.
The SFS algorithms assume that the surface reflectance
properties and the photometric model are known.
Unfortunately, the assumptions might be wrong or the
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
preconditions might not be available in practice. The
photometric model plays an important role in these problems.
In order to get the approximate photometric model, a probing
method may be appropriate. It means, one can approximate the
correct photometric model by using different known models as
probing models. The proper model will yield the least error
between the brightness image and the shaded image obtained
by using the probing photometric model. But the precondition
of this method needs a precision height image and the correct
photometric parameters (reflectance properties). As we see so
far, the difficulty is that one solution depends on another in a
circular problem. If we divide the de-shading task into three
sub-tasks: Determining Photometric Model (DPM),
Determining Reflectance Properties (DRP) and Determining
improved Height Image (DHI), each of these sub-tasks needs
the outputs of the other two as its inputs. This can be illustrated
as in Fig. 1.
Considering the above analysis, a de-shading system shown in
Fig. 2 is developed to perform the de-shading problem. It
contains two frames. One is the training frame, the other is the
working frame. The inputs to this system are the illumination
direction (7,7), a brightness image I(x,y), and an
1029
EM anti mn |
