OPTIMIZATION OF AUTOMATIC IMAGE REGISTRATION ALGORITHMS 
AND CHARACTERIZATION 
Ch.Venkateswara Rao *, Dr. K.M.M.Rao *, A.S.Manjunath *, R.V.N.Srinivas ^ 
" National Remote Sensing Agency, Hyderabad(rao cv, rao kmm, manjunath_as )@nrsa.gov.in 
b 
M.Tech project student from BIT-Mesra, Ranchi (srinivasrvn@rediffmail.com) 
PS Comission IH, WG,III/8 
KEY WORDS: Remote Sensing, Geometry, Medicine, Registration, Automation, Accuracy. 
ABSTRACT: 
In many image-processing applications it is necessary to register multiple images of the same scene acquired by different sensors, or 
images taken by the same sensor but at different times. Mathematical modeling techniques are used to correct the geometric errors 
like translation, scaling and rotation of the input image to that of the reference image, so that these images can be used in various 
applications like change detection, image fusion etc. In the conventional methods, these errors are corrected by taking control points 
over the image and these points are used to establish the mathematical model. The objective of this paper is to implement and 
evaluate a set of automatic registration algorithms to correct the geometric errors of the input image with respect to the reference 
image, by increasing the accuracy level of the registration and reducing the RMS error to less than a pixel. Various algorithms such 
as Wavelet transformation method, Fast Fourier transformation method, Morphological Pyramid approach and Genetic Algorithms 
are developed and compared. These algorithms are capable of considering the transformation model to sub-pixel accuracy. The 
benefits of these methods are accuracy, stability of estimation, automated solution and the low computational cost. 
1. INTRODUCTION 
[mage registration is one of the basic image processing 
operations in remote sensing. By registering the two different 
images acquired during different times or by different sensors 
can be used in various applications like change detection, 
image fusion (A.S.Kumar, 2003) etc. Most of image 
registration approaches fall into local or global methods. Local 
methods are referred to as rubber sheeting or the control-points 
method. Global methods involve finding a single 
transformation imposed on the whole image and are also 
referred to as automatic registration methods. Registration 
methods (L.G.Brown, 1992), can be viewed as different 
combinations of choices for the following four components: 
(1) Feature space 
(2) Search space 
(3) Search strategy and 
(4) Similarity metric. 
The Feature space extracts the information in the images that 
will be used for matching. The Search space is the class of 
transformations that is capable of aligning the images. The 
Search strategy decides how to choose the next transformation 
from this space, to be tested in the search for the optimal 
transformation. The Similarity metric determines the relative 
merit for each test. Search continues according to the search 
strategy until a transformation is found whose similarity 
measure is satisfactory. The types of variations present in the 
images will determine the selection. for each of these 
components. 
For example, the problem of registering two images taken of the 
same place at different times can be considered. Assuming that 
the primary difference in acquisition of the images was a small 
translation of the scanner, the search space might be a set of 
small translations. For each translation of the edges of the left 
image onto the edges of the right image, a measure of similarity 
would be computed. A typical similarity measure would be the 
correlation between the images. If the similarity measure is 
computed for all translations then the search strategy is simply 
exhaustive. The images are registered using the translation, 
698 
which optimizes the similarity criterion. However, the choice of 
using edges for features, translations for the search space, 
exhaustive search for the search strategy and correlation for the 
similarity metric will influence the outcome of this registration. 
In general all the image registration techniques evaluated during 
this study, were based on local methods that required manual 
selection of ground control points (GCPs) over the image and 
these points are used to establish the mathematical model .The 
selection of these control points is subjective and can lead to 
inconsistencies as it is interactive with the operator. The 
objective of this paper is to characterise a set of automatic 
registration algorithms to correct the geometric errors of the 
input image with respect to the reference image, by increasing 
the accuracy level of the registration and reducing the RMS 
error to less than a pixel. 
In the next section, we have provided a brief overview of some 
of the related work in this area. Sections 3 and 4 describe the 
methodology followed by the experimental results obtained on 
some common data sets. Lastly we have also provided some 
comparative measures on efficiency of various parameters 
between the different algorithms evaluated. 
2.AUTOMATIC REGISTRATION METHODS 
The image registration process is usually carried out in three 
steps (Leila M. Fonesca, 1997). The first step consists of 
selection of features. In the next step each of these features are 
compared with potential corresponding features of the other 
image. A pair of matched features is accepted as a control point 
(CP). Finally using these CPs a transformation is established to 
model the deformation between the images. To carry out this 
process automatically several algorithms have been proposed 
and were divided into the following classes (B.S.Reddy, 1996). 
(1) Algorithms that directly use image pixel values 
(2) Algorithms that operate in the frequency domain 
(3) Algorithms that use low-level features such as edges 
and corners and 
(4) Algorithms that use high-level features such as 
identified objects, or features. 
After studying various algorithms the following four methods 
    
    
    
   
   
     
    
   
   
   
    
  
  
   
    
   
  
    
   
   
      
   
       
      
   
    
   
    
   
    
    
  
  
    
    
    
     
    
    
    
    
    
   
   
    
  
  
  
Inte. 
wer 
Wa 
use: 
199 
con 
the 
ima 
leve 
mea 
feat 
tran 
erro 
ima 
mar 
ima 
ima 
detc 
Fas 
Yık 
doe: 
com 
dete 
inve 
app! 
dete 
thes 
COO! 
repr 
para 
the 
refe 
Mo 
(Zh 
dete 
radi 
ima; 
the 
mail 
opti 
para 
pixe 
inte; 
Ima 
Holl 
ensu 
cond 
pop 
chro 
rand 
crite 
their 
the | 
sub : 
Mat 
geor 
inpu 
warp 
calle 
the i 
a) 1 
b) 
diffe 
Each