430079,
n two aspects:
n illumination
on and uneven
ever, the main
t other factors.
nption of grey
ng is proposed
:ast effectively
pectra and the
imaging result,
ind radiometry
: mostly edge
ie presentation
en color, and
idation model.
u and Danaher
el based on the
>| based on the
ving geometry
2000; Chehdi,
Zhang, 2003a]
'estoration and
However, they
e definition of
ging lightness
But presently,
logy in image
ealistic 1mage
rpretation and
n attribution is
y and remote
| other areas.
ems above is
n II describes
| and indicates
ry and remote
d on physical
. Section IV
based on the
inalyzes their
| V proposes a
oncerned with
result.
2. BACKGROUND
It should be noticed that the image discussed here are all
captured in visible spectra bands, not relating to multi-spectra
or hyper-spectra.
For remote sensing image, the problem discussed above is
relative to radiometry correction closely all the while. The task
of radiometry correction is to build the quantitative relation
between the signal quanta of each unit of the sensor and the
actual ground object radiometry quanta the pixel correspond to
the unit, develop the correction coefficient of the sensor directly,
and realize the radiometry correction on the acquired data.
Based on the causes of radiometry distortion, the radiometry
correction can be performed in three directions. They are
atmosphere correction, sensor correction and the correction for
the distortion formed by the solar altitude angle and terrain
affection. In physical sense, build a model that describes main
elements in the imaging process can accurately and effectively
restore the radiometry characteristic of the ground undoubtedly.
A lot of researches have been performed on this topic, which
includes building ground radiometry field [Zhang, 1995],
atmosphere radiometry correction based on image [Tian, 1998],
radiometry correction based on terrain [Yan, 2000; Li, 1995],
and so on. It’s obvious that the ideal atmosphere radiometry
correction and reflectance inversion should only make use of
image information, instead of measuring support data in fields,
and is practicable for history data and desolate area. Hence,
although restore the surface spectra characteristic in the scene
accurately should build a precise and complete model for
quantitative remote sensing, for a mass of image analysis and
applications such as photo interpretation, 3-D reconstruction,
and feature extraction, accomplish the task of radiometry
correction conveniently and effectively is still significant.
Additional, for aero-borne image and close range image, the
image degradation may directly affect the applications such as
visual inspection, image retrieve, texture mapping and mosaic
etc.. Because the imaging scale is relative smaller than remote
sensing image, this two kinds of images are affected less by
atmosphere relatively, but more by the geometry structure of
the imaging scene. Therefore, building an illumination model to
take into account all factors in the imaging process completely
is difficult. Additional, with the improvement of the resolution
of the sensors, the differences on resolution between kinds of
scales become not significant any more. So, building a model or
application framework only based on the image information
and is available in most images is significant.
Therefore, in the view of data type and affection factor,
processing the affection on imaging scene formed by
illumination environment usually has two directions. The first is
to build a model. The second is to restore based on image
information. And the latter is the main problem we considered.
With the development of image capturing device, displaying
device and output device, the application request on image
realistic reproduction is improved too. So, the definition of
dodging in photogrammetry and remote sensing should be
extended to include color restoration, which means the problem
dodging processed includes uneven lightness, uneven color,
obvious color cast, realistic reproduction and geometry
degradation to obtain realistic image suitable to human vision
Psychology feature and physical feature, instead of
conventional uneven lightness.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BS. Istanbul 2004
3. DODGING BASED ON ILLUMINATION MODEL
Strictly speaking, the approaches based on physical model can
deal with not only the dodging problem. In areas such as
computer vision, machine vision, computer graphics, etc.,
building an illumination model can simulate the illumination
environment to re-rendition the scene, including dodging and
obtaining the representations in various illumination
environments. So far, literatures indicate that the building of
illumination model can performed in three directions:
conventional illumination model, complete plenoptic function
without illumination model, and image based rendering. [Shen,
2000]
3.1 Conventional Illumination Model
After obtained the description on illuminant, surface reflectance
in the scene, transitive characteristic of sensors, and geometry
relation of the three elements above, builds a physical model to
modify the color of each pixel. That is the basic framework of
conventional illumination model [Fournier, 1993; Shum, 2000;
Kang, 1999]. Some literatures [Wu, 2001; Danaher, 2001] have
brought such method to bear on the radiometry correction on
remote sensing data and achieved excellent result. Because the
illuminant is mostly the sun, and supposing the ground is
Lambertian, the primary task is to computer the bi-directional
reflectance distribution function (BRDF). In most applications,
take use of conventional illumination model should sample
some ground objects! spectra distribution. Because such
function is a 5-D function (including wavelength), the sampling
and computation are tedious. Although the literature [Shen,
2000] indicates that obtain some BRDF sample of the surface
without practical measurement in the field is possible, such as
calculates pseudo-BRDF of the surfaces based on the sky light
[Yu, 1998], calculate day light spectra from the illumination
color temperature based on empirical data [Takagi, 1990],
extracts. BRDF from the reflect nature of the surface
[Jaroszhiewiez, 2003], etc., whether the reflectance data is
measured or not, to restore the BRDF of surface in the scene is
still a challenging problem. For the applications in
photogrammetry and remote sensing, such model can be
simplified because the illuminant is ideal and only one - the sun.
In remote sensing and aero-photogrammetry, because the
objects and terrain affection are various and complex, surfaces,
usually is the ground, can be looked as Lambertian, while in
close range photogrammetry, the surface is usually the building
surface or other special objects, they can't be looked as
Lambertian. Therefore, building a conventional illumination
model is an effective method in remote sensing image
processing with reflectance data measured in field. However,
it’s difficult in processing image in remote sensing without
measured data and in close range applications, which make it be
researched in depth in future to obtain practical method.
3.2 Plenoptic Function without Illumination Model
Because to restore the whole scene is a very difficult business,
some researchers attempt to build a model to avoid the complex
computation in the model building through controlling the
variables in the scene and sampling densely. Such attempts take
full advantage of the continuity of the plenoptic function. If the
variables in the scene are looked as variables in the plenoptic
function, a description of plenoptic function in a higher
dimension is obtained. Such method without illumination model
is equal to restore the continuous expression of that high
dimension plenoptic through dense discrete sampling to predict
