Analysis of Feature Detection Method in Image Registration

I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com

Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 9 7

Abstract: Image registration is the process of transforming
the different sets of data into one coordinate system.
Registration is necessary in order to be able to compare or
integrate the data obtained from different measurements such
as different viewpoints, different times, different sensors etc.
Image Registration is a crucial step in all image analysis
tasks in which the final information is gained from the
combination of various data sources, change detection, and
multichannel image restoration. One of the applications of
image registration is Corner detection. Corners determine the
contours characteristics of the target image, and the number
of corners is far smaller than the number of image pixels,
thus can be a good feature for image registration. By
considering the algorithm speed and registration accuracy of
the image registration, we can make the registration more
efficient and for that Harris Corner detection and other
methods are used. In this paper performance analysis
performed will be: registration speed, MSE (mean square
error) and quality of registration.
Keywords: Corner detection, Feature selection,
transformation model estimation, corner pairs, Root
Mean Square Error, Quality.

1. INTRODUCTION
Registration is a fundamental task in image processing
used to match two or more pictures taken, for example, at
different times, from different sensors, or from different
viewpoints. Image Registration is a crucial step in all
image analysis tasks in which the final information is
gained from the combination of various data sources,
change detection, and multichannel image restoration .It
geometrically aligns two imagesthe reference and
sensed images. One of the applications of image
registration is Corner detection. Corners determine the
contours characteristics of the target image, and the
number of corners is far smaller than the number of
image pixels, thus can be a good feature for image
registration [1].Registration methods divided in to two
following categories:
Region based methods: adopts correlation technique and
are easy to calculate and implement but required
large amount of calculations which is a serious
problem[4].
Features based methods: extracts some common
invariant features such as corners, contour etc. Corners
have the merits of the rotation invariance and almost
immunity to illumination conditions. It is the widely used
method in image registration process[5].

2. STEPS INVOLVED IN IMAGE
REGISTRATION
Image registration essentially consists of following steps:
2.1 Feature detection
A feature is defined as an "interesting" part of an image.
Salient and distinctive objects (closed-boundary regions,
edges, contours, line intersections, corners, etc) in both
reference and sensed image are detected. For further
processing, these features can be represented by their
point representatives (centers of gravity, line endings,
distinctive points), which are called control points (CPs).
The features represent information on higher level [4].
2.2 Feature matching
The correspondence between the features in the reference
and sensed image established. The detected features in the
reference and sensed images can be matched by means of
the image intensity values in their close neighborhoods,
the feature spatial distribution, or the feature symbolic
description. Some methods, while looking for the feature
correspondence, simultaneously estimate the parameters
of mapping functions and thus merge the second and
third registration steps. The two major categories (area-
based and feature-based methods, respectively), are
retained and further classified into subcategories
according to the basic ideas of the matching methods[6].
2.3 Transform model estimation
The type and parameters of the so-called mapping
functions, aligning the sensed image with
the reference image, are estimated. After the feature
correspondence has been established the mapping
function is constructed. It should transform the sensed
image to overlay it over the reference one. The
correspondence of the CPs from the sensed and reference
images together with the fact that the corresponding CP
pairs should be as close as possible after the sensed image
transformation are employed in the mapping function
design. The task to be solved consists of choosing the type
of the mapping function and its parameter estimation.
The type of the mapping function should correspond to
the assumed geometric deformation of the sensed image,
to the method of image acquisition (e.g. scanner
dependent distortions and errors) and to the required
accuracy of the registration [5].
2.4 Image resampling and transformation
The sensed image is transformed by means of the
mapping functions. The mapping functions constructed
during the previous step are used to transform the sensed
image and thus to register the images. The transformation
Analysis of feature detection method in Image
Registration

Rajvinder Kaur

NITTTR, Sec.26, Chandigarh, India


can be realized in a forward or backward manner. Each
pixel from the sensed image can be directly transformed
using the estimated mapping functions. This approach,
called a forward method, is complicated to implement, as
it can produce holes and/or overlaps in the output image.
Hence, the backward approach is usually sensed image
on the regular grid. In this way neither holes nor overlaps
can occur in the output image [5].
3. IMAGE REGISTRATION
METHODOLOGY
The translational image registration problem can be
characterized as follows: We are given functions F(x) and
G(x) which give the respective pixel value at each
location x in two images, where x is a vector. We wish to
find the disparity vector h that minimizes some measures
of the difference between F(x+h) and G(x), for x in some
region of interest R.

Figure 1 The image registration problem

Image registration is an important part of the image
processing and computer vision. Image registration
methods usually divided into two categories: region-
based methods and feature-based methods. The former
adopts correlation technique to determine corresponding
matching position between two images. it is easy

to
implement, but large amount of calculation is serious
problem,

especially ,due to rotation, scaling and shifting
transformation influence and distortion such as shading,
degradation of images, the algorithm effectiveness will be
dropped dramatically; the latter firstly extracts some
common invariant features, such as contour, moments,
etc., next, fine matching is conducted, because of
invariant features, nothing to do with the image gray
level, the ambiguity of registration can be effectively
solved [1].
The commonly used automatic point feature extraction
algorithms include: the method based on wavelet
transform, SUSAN operator, Harris operator, Forstner
operator, Morave operator and so on. Harris operator has
been proved to be one of the best corner detection
operators[3].
3.1 Harris corner detection algorithm
Harris operator was put forward by C. Harris and M. J.
Stephens in 1988, which extracted point feature by the
first order image gradient. The method calculates square
matrix of the average gradient of each pixel, and then
feature points can be extracted by analyzing eigen value.
Experiment shows that feature points by this method
distribute evenly and reasonably, may extract feature
points quantitatively. Inadequately, the algorithm is only
a single scale corner detection, the result relies entirely on
the threshold value setting, if the threshold value is large,
it result in corner missing, if smaller threshold, also
easily mixed with false corner[2].Multi-scale space
representation of the image obtained by smoothing can be
described as (x,o) space, here, x and o is location
parameters and the scale parameter respective. When the
same images are filtered by using smoothing function of
different scales, the result images are a cluster of scale
space the original image relative to the smooth function,
o is the scale-space coordinates.
2
1
( , ) exp[ / (2 )]
2
G x x
=
Multi-scale Harris corner detection operator is put
forward, which transforms image into different scale
space, with the advantage of integrating small scale
precision positioning points with large-scale eliminating
the false points and retaining the true points. Specific
steps of multi scale Harris corner detection are as follows

(a) The scale space expression of Harris operator. The
corner measurement of different scales and scale change
are in accordance with the standard deviation o
i
= 0.5 : 5
of the Gaussian window, step length is 0.5, namely i =
1:10 , the calculation of the following formula forms the
matrix M of different scales
2 2
( ) ( ) ( )
x i i
I
I G
x
= -------------- (1)
2 2
( ) ( ) ( )
y i i
I
I G
y
= ---------------- (2)
2
( ) ( ) ( . )
xy i i
I I
I G
x y

= ------------- (3)
2
2
( ) ( )
( )
( ) ( )
x i xy i
i
xy i y i
I I
M
I I

| |
=
|
|
\ .
------------ (4)
Where, is the first derivative in the x direction, is
the first order derivative in the y direction, G(o) is
Gaussian function, M(o
i
) is matrix M in the scale o
i.

Corner response function is calculated in the scale
2
: ( ) det( ( )) 0.04*( ( ( )))
i i i i
R M trace M = -
---------- (5)
(b) For each scale, the response points of local maximum
are obtained by using non-maxima suppression method in
the same window of Gaussian kernel size. The former
C
num
is selected as candidate corner according to
descending order same window of Gaussian kernel size. If
the minimum scale point is only included in the window,


the point is deleted. Thus the point can be ensured to
achieve in the scale as small as possible, the positioning
accuracy is further improved, the information point in
large scale is still retained, meanwhile, deleting false
corner in small scale [1].
c) Sub-pixel accurate positioning of point. The process
can be accomplished by Taylor expansion of space scale
function D(x, y,o) in the local extreme point (x
0
, y
0
,o).
The multi-scale Harris corner detection method can
acquire corner information in a number of scales, corner
quantity is reasonable, the threshold in each scale neednt
be manually set, thus, reducing the threshold constraints
on the corner detection, realizing corner accurate
positioning in the small scale and ensuring the false
corner eliminated and the true corner retained in the large
scale[9].
Set
1
and
2
as the two eigen values of M.
1
and
2
are
the rotation invariants of M, and proportional to the
principal curvatures of the local autocorrelation function.
At this point, the flat areas, corners and edges can be
judged by the value feature of
1
and
2,
' there are three
conditions:
(1) Both of two curvatures are small, there implies a flat
area;
(2) One is large, the other is small. It suggests that the
partial autocorrelation function is ridge-like, the change
along the ridge is little, while the change perpendicular to
the ridge is great, there implies an edge.
(3) Both of the curvatures are large. There implies a peak,
and the change along any direction is very intense,
indicates the current point is a corner.
Modified Harris corner detection include following steps:
Step I: Operate each pixel on an image, find out the new
calculated value of the matrix M, and estimate the comer
score of the corresponding pixel.
Step 2: Partition the image into a fixed-number of blocks,
detect the comers for each image block, then store and
sort the detected comer score in the array Max[S] (S is the
number of comer points).
Step 3: Select the top-n comer points in Max[S] as the
candidate comer points. In order to ensure that all sub-
blocks have reserved comers, here a special policy is used
to control the iteration. First, an initial value of k is
selected from range[0, I]. Then, using k=k+s (s is a step
length) to iterate until k=l, and determine whether every
image block has comer points or not. Finally, remove the
adjacent comers for the entire image.
3.2 Affine transformation parameter estimation
After determining the registration points, it needs to fit
the six best parameters of affine transformation using
least square method. Assuming the selected control points
sequence in reference image is (x
1
, y
1
), (x
2
, y
2
), (x
n
, y
n
),
(x
1
,y
1
) , (X
2
, y
2
,) ,., (x
n
, y
n
) is the sequence that of the
unregistered image. If is used to represent the matrix, the
least squares solution of six parameter of affine
transformation is described as follows,

00 10
01 11
[ 1, 2,.... ] , [ 1, 2,..... ]
T T
x t
a a
a P x x xn a P y y yn
a t
+ +
| | | |
| |
= =
| |
| |
\ . \ .

Where, the P
+
is the generalized inverse matrix of the P
[1].

3.3 Resampling
After determining the transformation parameters, the
registration image needs to be re-sampled. There are two
models for re-sampling, one is the direct sampling, and
the other is the indirect sampling.

Figure 2 Direct and indirect re-sampling diagram

As the computer raster image is discrete, if using the
direct re-sampling model, there is no way to guarantee
that the pixels of original image can just be mapped to the
matching pixels of the registered image, therefore we
exploit the indirect re-sampling model. In other words, it
starts from the registered image to find a corresponding
position on the original image, and then estimates the
gray value of current position by interpolating the gray
value of the current neighborhood, and finally reassigns
the obtained gray value to the position locates in the
registered image[1].

4. ALLORITHM USED
Step I: Images are taken for demonstration.
StepII: Feature detection algorithm is applied on images
[1]. There are various feature detection algorithms but in
this work of image registration, Harris Corner Detection
method has been used. This method has following
characteristics:
(a) Harris Operator has been proved to be one of the best
corner detection algorithms [2].
(b)The method extracts point features by the first order
image gradient.
(c) It calculates square matrix of the average gradient of
each pixel.
(d)Then the feature points are extracted by analyzing the
values.
Step III: Calculation of transformation parameters is done
in this step: Various affine transformation parameters can
be calculated by the coordinates of control points.
StepIV: Affine Transformation is applied: In this step,
affine transformation is applied to the part of the image.
Step V: Resampling- After determining the
transformation parameters, the registration image needs
to be re- sampled [2].
Step VI: Output the final registered image.


Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 1 0 0

5. RESULT

Figure 3 Input image

Figure 4 Reference image

Figure 5 Registered image

Table1: Index Values of above registration experiment

INDE
X
RMSE ENTROP
Y
SD QUALI
TY
Aeria
l
Image
0.2577
2
15.4289 54.798
1
1.0854

6. CONCLUSION
On this basis, to solve the research problems, comer
detection method based on Harris model is proposed and
applied for automatic Image registration. This method
effectively eliminate features points that have false
correspondence, thus the comer points detected distribute
more reasonably, and the image registration become
faster. The experiments also showed the effect of image
registration is Satisfactory.
The results show that aerial images and satellite images
achieve a very good automatic registration by adopting
affine transformation. Compared with the traditional
image registration methods, it is simple, low
computational complexity, no manual operation under
optimal parameters conditions, stable and reliable.
In future modified methods of feature detection will be
implemented.

REFERENCES
[1] Yijian Pei, Hao Wu, Jiang Yu, and Guanghui Cai,
Effective image registration based on improved
harris corner detection, IEEE International
Conference on Information, Networking and
Automation (ICINA) ,Volume.1, pp. V1-93 - V1-96,
2010.
[2] J.B. Ryu, C.-G. Lee and H.-H. Park, Formula for
Harris corner detector, IEEE electronic letters,
Volume.47, No. 3, 3rd February 2011.
[3] J.B. Ryu, H.-H. Park and J. Park, Corner
classification using Harris algorithm, IEEE
electronic letters, Volume.47, No. 9 , 28th April
2011.
[4] Hui Lin, Peijun Du, Weichang Zhao, Lianpeng
Zhang and Huasheng Sun, Image registration based
on corner detection and affine transformation, IEEE
International Congress on Image and Signal
Processing , Volume.5, pp. 2184 - 2188 ,2010.
[5] J. Salvi, e. Matabosch, D. Fofi, and J. Forest, "A
review of recent range image registration methods
with accuracy evaluation," Image and Vision
Computing, Vo1ume.25, No.5, 2007, pp.578-596,
doi: 10. 1 016/j.imavis.2006.05.012.
[6]Barbara Zitova and Jan Flusser, Image registration
methods: a survey, International Journals of Image
and Vision Computing, pp. 9771000, 2003.
[7]F. Bellavia D. Tegolo C. Valenti, Improving Harris
corner selection strategy, IET Comput. Vis., 2011,
Volume.5, Issue. 2, pp. 8796.
[8]Edward Rosten, Reid Porter, and Tom Drummond,
Faster and Better: A Machine Learning Approach to
Corner Detection, IEEE transactions on pattern
analysis and machine intelligence, Volume. 32, No.
1, January 2010.
[9]Medha V. Wyawahare, Dr. Pradeep M. Patil,
Hemant and K. Abhyankar, Image Registration
Techniques: An overview, International Journal of
Signal Processing, Image Processing and Pattern
Recognition, Volume. 2, No.3, September 2009.
[10]Pratt, W.K. Correlation Techniques of Image
Registration, IEEE Journals & Magazines,
Volume:AES-10, Issue.3 , pp. 353 358,1974.

Rajvinder Kaur, born in 1986, received
her B.Eng. honours degree in Computer
Science in 2004 from Kurkshetra
University, and pursuing M.Tech degree
from NITTTR Chandigarh. Currently she
is working as Assistant Professor in CSE
department at SPGC Lalru.

Analysis of Feature Detection Method in Image Registration

Uploaded by

Copyright:

Available Formats

Analysis of Feature Detection Method in Image Registration

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Analysis of Feature Detection Method in Image Registration

Uploaded by

Copyright:

Available Formats

I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)

Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com

You might also like