International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
choice for varying radiometric and geometric images
(Section 2).
e The appropriate registration transformation function is not
investigated (ie, simplified and sometimes invalid
registration transformation function is assumed).
e The developed similarity measures for matching primitives
are empirical and sometimes subjective. Cross-correlation
and least squares matching are the best known criteria to
compare the degree of similarity. Here, the images to be
matched have to be radiometrically very similar, preferably
imaged by the same sensor. However, gray level
characteristics of the images to be matched can vary from
sensor to sensor and hence correlation measures become
unreliable (Fonseca and Manjunath, 1996). Moreover,
applying cross-correlation requires two images with same
resolution which disagree with existing satellite images (i.e.,
IKONOS (1m), SPOT (10m), LANDSAT (30m), etc.)
Prior methods have certain advantages in computing the
transformation parameters in a single step and in retaining the
traditional way of thinking about registration in the sense of
identifying similar features first and then computing the
parameters of the registration transformation function. The
suggested approach significantly differs from the other
registration strategies as it uses straight lines features for
simultaneously determining the correspondences between the
involved primitives and solving for the parameters of the
registration transformation function.
This paper outlines a comprehensive image registration
paradigm that can handle multi-source imagery with varying
geometric and radiometric properties. The most appropriate
primitives (Section 2), transformation function (Section 3), and
similarity measure (Section 4) has been incorporated in a
matching strategy (Section 5) to solve the registration problem.
Experimental results using real data proved the feasibility and
the robustness of the suggested paradigm are discussed in
Section 6. Finally conclusions and remarks are drawn in
Section 7.
2. REGISTRATION PRIMITIVES
The registration primitives encompass the domain in which
information is extracted from input imagery for the registration
process, mainly: distinct points, linear features, and
homogenous/areal regions, Figure 1.
Distinct'points Linear features Areal regions
Figure 1. Alternatives of registration primitives
2.1 Points
Traditional procedures for manually registering an image pair
require interactive selection of tie points in each image. Such
tie points are then used to determine the parameters of a
registration transformation function, which is subsequently used
to resample one of the images into the reference frame
.associated with the other image. However, such a procedure can
928
lead to inaccurate results and is slow to execute, especially if a
large number of images with varying geometric and radiometric
properties need to be registered. Visually inspecting the
imagery, one can see that manual identification of conjugate
points is extremely difficult if not impossible, Figure |.
Automatic extraction of points based on the radiometric
information results in different sets of points from each image
due to varying radiometric properties of involved imagery. This
situation extends to the problem of finding conjugate points
where it would be unlikely that point extraction algorithms
would be able to identify the same point. In other words, for
multi-source imagery with varying geometric and radiometric
resolutions, the texture and gray levels at the location of
conjugate points will not be similar. Therefore, automatically
and/or manually extracted points will be difficult to match and
are not suitable primitives for registration. Consequently, linear
and areal features will be considered and investigated for its
suitability for multi-source image registration since the
geometric distribution of the pixels making up the feature can
be used in the matching, rather than their radiometric attributes.
2.2 Linear features
In contrast to point primitives, linear features have a set of
appealing properties when they appear on multi resolution
images especially in urban areas. These properties include the
following facts:
e Compared to distinct points, linear features have higher
semantics, which can be useful for subsequent processes
(such as DEM generation, map compilation, change
detection, and object recognition).
e Images of man-made environment are rich with linear
features.
e |t is easier to automatically extract linear features from
imagery rather than distinct points (Kubik, 1991).
e Geometric constraints are more likely to exist among linear
features. This can lead to a simple and robust registration
procedure.
2.3 Areal Features
Areal primitives might not always be available especially when
dealing with satellite scenes over urban areas. Moreover,
registration procedures based on areal primitives use the centers
of gravity of these features as the registration primitives. The
estimated centers of gravity are susceptible to potential errors
associated with the identified boundaries of these patches.
Compared to linear features, areal features are less appropriate
considering availability in nature, complexity of extraction
algorithms, and existence of geometric constraints. Areal
features can be represented as a sequence of linear features
through the replacement of its boundaries.
Based on the above analysis of different candidate primitives,
this paper will adopt the linear features in the registration
process. Straight lines, a subset of linear features, possess
further attracting benefits that made it the premium choice as
explained in the following subsection.
2.4 Straight Lines
Linear features can be represented either by an analytical
function (e.g., straight lines, conic sections, or parametric
Int
lin
be
va
rer
lin
At
fur
CO!
inf
rel
sp:
ref
tas
the
vie
tec
tra
sp
cir
Th
col
ma
poi
nai
LA
An
gai
wit
op
20(
the
RF
Sce
de
the
For
ang
ma
coc
CoC
eig
var
