A Geometric and Radiometric Simultaneous
Correction Model (GRSCM) Framework for High-
Accuracy Remotely Sensed Image Preprocessing
Chang Li and Hao Xiong
Abstract
The grey value g (x, y) of pixel on radiometric spectrum is
regarded as a function of the geometric coordinates (x, y).
Hence, there is a unity of opposite relationships between the
geometric and radiometric information, such that, these two
types of information cannot be separated. Therefore, this pa-
per proposes a novel geometric and radiometric simultaneous
correction model (
GRSCM
) framework inspired and developed
from least squares matching (
LSM
). Based on the Gauss-Mar-
kov model, geometric and radiometric correction coefficients
are integrated and solved by an iterative method with variable
weights in the proposed model. Moreover, many state-of-the-
art models and methods can be integrated into the proposed
general
GRSCM
framework. In the
GRSCM
of this paper, RAN-
dom SAmple Consensus (
RANSAC
), stepwise regression and
significance testing are integrated and used. The experimental
results demonstrate that the accuracy of the
GRSCM
is signifi-
cantly improved compared with that of geometric correction
and radiometric correction separately.
Introduction
Raw remotely sensed images always contain geometric
and radiometric distortions originating primarily from the
remotely-sensed platform, the sensor, the atmosphere and the
Earth (Jensen, 1996). Such images cannot be used directly in
a geographic information system (
GIS
) (Toutin 2004). Con-
sequently, one important procedure prior to using remotely
sensed images is geometric and radiometric correction. How-
ever, traditional methods have been processing geometric and
radiometric corrections, respectively, (Queiroz-Neto
et al
.,
2004), which can be illustrated by the following.
The main purpose of geometric correction or rectification
is to explicitly determine the mapping function using ground
control points (
GCPs
) or ground control lines (GCLs) and to
determine the pixel brightness in the corrected image (Jensen,
1996; Devaraj and Shah, 2014). Geometric correction can also
be divided into two models: (a) rigorous sensor models (i.e.,
physical model) including coplanar equation and collinearity
equation; and (b) non-rigorous models, e.g., polynomial mod-
els, and rational polynomial coefficient (
RPC
), namely rational
function model (
RFM
). However, the polynomial model is
usually used more widely (Shi and Shaker, 2003), because it
is simpler than physical model without a lot of known condi-
tions. Many previous studies have indicated that the geomet-
ric accuracy of the corrected image is affected by the number,
precision, and spatial pattern of
GCPs
(Orti, 1981; Labovitz and
Marvin, 1986; Mather, 1995; Zhou and Li, 2000; Wang
et al
.,
2005; Sertel
et al
., 2007; Radhadevi, 2013). Besides
GCPs
, GCLs
are another important image feature that are often employed
in image correction, e.g., generalized point photogrammetry
(
GPP
) (Zhang and Zhang, 2004; Zhang
et al
., 2008.) and a gen-
eralized-line-based iterative transformation model (GLBITM)
(Li and Shi, 2014). However, in current image preprocessing,
the geometric correction algorithm does not consider the ra-
diometric distortion of images. In automatic image matching,
the radiometric difference cannot be ignored as noise, and it
has a significant effect on image matching to acquire conju-
gate points that can be also used as
GCPs
. This means that the
relationship between geometric and radiometric distortion
is the unity of opposites, and the geometric and radiometric
correction should therefore be simultaneously processed in a
scientific method. Moreover, when the geometric correction is
finished using the classic method, its coefficients will not be
changed in subsequent steps. Therefore, this study proposes
that, after the initial geometric and radiometric coefficients
are obtained, these coefficients will be further solved and
optimized together simultaneously in the subsequent correc-
tion. As a consequence, the geometric correction should be
repeated in the optimization processing.
Radiometric correction or calibration is used to correct
the radiometric distortion of remotely sensed images. It can
be divided into two different methods: absolute radiometric
correction and relative radiometric correction. Relative radio-
metric correction is often more commonly used to eliminate
the radiometric distortion of images, because it is difficult to
implement absolute radiometric correction in most real-world
situations. Actually, a number of methods using linear regres-
sions have been tried for relative radiometric correction. In
this implementation, the critical aspect is the determination
of suitable time-invariant features upon which to base the cor-
rection (Schott et al., 1988; Strebel et al., 1991; Moran, Jack-
son et al., 1992; Yang and Lo, 2000; Du Teillet
et al
., 2002). A
completely automatic procedure was suggested to determine
time-invariant observations (Canty et al., 2004; Nielsen, 2007;
Canty and Nielsen, 2008; Li
et al
., 2016). In the studies cited
above, the geometric information is not always considered
in linear relative radiometric correction. However, the result
of geometric registration affects the corresponding pixels.
Because image registration inevitably contains error, the equa-
tion of linear radiometric correction, which is based on the
corresponding pixels, is not accurate so as to lead to the prob-
lem of remotely sensed application (e.g., change detection).
Furthermore, the relationship between geometric and radio-
metric information is the unity of opposites. Consequently,
simultaneous geometric and radiometric processing is benefi-
cial for improving correction accuracy.
Key Laboratory for Geographical Process Analysis &
Simulation, Hubei Province, and College of Urban and
Environmental Science, Central China Normal University,
Wuhan 430079,
,
).
Photogrammetric Engineering & Remote Sensing
Vol. 83, No. 9, September 2017, pp. 621–632.
0099-1112/17/621–632
© 2017 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.83.9.621
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
September 2017
621
