| 2004 
——— 
ember 
tching 
2000, 
2) 4A 
nent”. 
'emote 
uction 
Aodels 
‘Using 
zes by 
ument 
d 'K, 
ngs of 
Tence. 
itm. 
atzara, 
ge in a 
ol. 11 
COMPREHENSIVE PARADIGM FOR SEMI- AUTOMATIC REGISTRATION OF 
MULTI-SOURCE IMAGERY 
R. I. AL-Ruzouq 
Department of Geomatics Engineering, University of Calgary 
2500, University Drive NW, Calgary AB T2N 1N4 Canada - al-ruzoug@geomatics.ucalgary.ca 
PS ThS 1, Integration and Fusion of Data and Models 
KEY WORDS: Image Registration, Features, Automation, Matching, Multi-Resolution, Transformation 
ABSTRACT: 
The enormous increase in the volume of remotely sensed data, which might be in different formats and relative to different reference 
frames, has created the need for robust data processing techniques that can fuse data observed by different acquisition systems. This 
need is motivated by the fact that collected data by these sensors are complementary in nature. Therefore, simultaneous utilization 
of the collected data would guarantee full understanding of the object/phenomenon under consideration. In this regard, a registration 
procedure can be defined as being concerned with the problem of how to combine data and/or information from multiple sensors in 
order to achieve improved accuracies and better inference about the environment than could be attained through the use of a single 
sensor. Registration of multi-source imagery captured under different conditions is a challenging problem. The difficultv is 
attributed to the varying radiometric and geometric resolutions of the acquired imagery. In general, an automatic image registration 
methodology must deal with four issues; registration primitives, transformation function, similarity measure and matching strategy. 
This paper outlines a comprehensive image registration paradigm that can handle multi-source imagery with varying geometric and 
radiometric properties. The most appropriate primitives, transformation function, and similarity measure have been incorporated in a 
matching strategy to solve the registration problem. Experimental results using real data proved the feasibility and the robustness of 
the suggested paradigm. 
1. INTRODUCTION Automatic and even manual registration of imagery remains 
challenging for several reasons. First of all, imagery and/or data 
Image registration is concerned with the problem of how to sets are usually acquired using different sensor types, each 
combine data and/or information from multiple sensors in order having its inherent noise. Furthermore, radiometric as well as 
to achieve improved accuracies and better inference about the geometric properties of the same object in the involved imagery 
environment than could be attained through the use of a single might differ as a result of changes in the sensor view point, 
sensor. In some applications, image registration is the final goal imaging methodology, imaging conditions (e.g., atmospheric 
(e.g., interactive remote sensing, medical imaging, etc.) and in changes, cloud coverage, and shadows), and spectral sensitivity 
others, it is a prerequisite for accomplishing high-level tasks of the involved imaging systems (e.g., panchromatic, multi- and 
such as sensor fusion, surface reconstruction, and object hyper-spectral imaging systems). Finally, the registration 
recognition. With the flux of high resolution scenes captured by process can be complicated by changes in object space caused 
space-borne — platforms (eg, LANDSAT-7, | IKONOS, by movements, deformations, and urban development between 
QUICKBIRD, ORBVIEW, EROS-AI, and SPOT-5), there is an the epochs of capture associated with the involved images. This 
Increasing need for a robust registration technique, which can paper will investigate and develop a semi-automated, accurate, 
tolerate varying geometric resolutions of the available scenes. and robust registration paradigm that can cope with the 
abovementioned challenges and problems. 
In general, an automatic image registration methodology must 
deal with four issues. First, a decision has to be made regarding Although there has been a vast body of research that has dealt 
the choice of the registration primitives, which refers to the with automatic image registration (Seedahmed and Martucci, 
features that will be extracted in the input imagery to solve the 2002; Dare and Dowman, 2001; Fonseca and Costa, 1997; 
registration problem. The second issue is concerned with Hsieh et al., 1997; Boardman et al., 1996; Flusser, 1992 and 
establishing the registration transformation function that Wolfson, 1990), we still do not have a methodology that meets 
mathematically describes the necessary transformation for the the current challenges posed by image registration. Drawbacks 
alignment of the images to be registered. Then, a similarity can be summarized by the following remarks: 
measure should be devised to describe the necessary constraints 
ensuring the correspondence of conjugate primitives. Finally, a 
matching strategy has to be designed and implemented as a 
controlling framework that utilizes the  primitives, the 
transformation function, and the similarity measure to solve the 
registration problem (ie. automatically determines the 
correspondences among conjugate primitives). 
e Points are usually used as a registration primitive. Even some 
techniques refer to regions and lines features such as lakes, 
rivers, cost-lines and roads. However, each of these features 
will be assigned one or more point locations (e.g. centroid of 
area, line endings, etc.) to be used as registration primitive 
(Fonseca and Manjunath, 1996). Points are not reliable 
927