ew of assessing
s of remotely
f Environment,
K. D. Helava,
y workstations.
"^ & Remote
' Digial Image
ig Perspective.
U.S.A., pp.
i, 1995. Image
G algorithm.
remote sensing,
Seiler, 1992. On
ge compression.
Compression
ymputer Society,
handani, 1992.
JPEG baseline
ages. IEEE, pp.
Mitchell, 1988.
)-coder. IBM J.
EG still picture
Transaction on
)p. 18-34.
Automatic stereo image matching using edge detection technique
Dr. Rehab H. Alwan,Mr. Mohamed A. Naji
Space Research center,Baghdad-IRAQ
Commission II WG /4
KEYWORDS: Stereo image, edge detection,edge matching,edge linking
ABSTRACT
An edge matching technique has been used in this work where an algorithm was developed for detecting &
coding the edge points which depends on the direction of the edge points the coding is followed by edge thining ,
edge linking & isolated points removing .Then points tracing process is performed to form the straight lines .
Lines matching operation is performed to group the lines in corresponding lines pairs .Features that are used in
the matching process are ‚line length ‚line orientation ends point coordinates & line location .
1. Introduction
Edge-based matching is the process in which two
representatives (edge) of the same object are pared
together .Any edge or its representation on one im-
age has to be compared and evaluated against all the
edges on the other image .The matching is basically
a selection process in which edges are pared accord-
ing to some measures of similarity .
In order to make the edges in a form suitable for
computer processing (especially stereo matching) it
is usually of interest to use edge representations and
descriptions .A lot of representations and approxima-
tions were proposed ,among these are :line represen-
tation (McIntosh,1988) ,polygon approximation
(Greenfelot,1989) ,
»-5- representation (toth C.1992), Hough representa-
tion and so many others. Special interest will be fo-
cused here on the line approximation and matching.
2. Lines Exrtaction
Line approximation techniques have been studied
since the early days of edge representations and ap-
proximations, and are described in text books
(Gonzales R. and Winlz P., 1987) ( Ballard .1982) .
Lines are commonly defined as collection of local
edges that are contiguous in the image. Thus most of
the algorithms rely on a two steps process for line
extraction: detection of edges using one of the edge
detection operators and then approximate these edg-
es to lines.
2.1. Point Detection
The problem of detecting isolated points in an image
applies in noise removal and particle analysis. The
basic mask used for detecting isolated points in an
image is shown in Fig.1.
1 „von
À Sl as]
E aS eI
Figure 1 : The mask used for detection isolated point
The center of this mask is moved from pixel to pixel
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
and convolved with the previous mask.
In an area of constant gray level the result of the op-
eration given above would be zero. On the other
hand, if the mask is centered at an isolated point in-
tensity is greater ((or less)) than the background then
the result would be greater ((less)) than zero.
2.2. Edge Detection
Change or discontinuities in image attribute such as
luminance and texture are fundamentally important
primitive features of an image since they often pro-
vide an indication of the physical extent of objects
within the image (Pratt .1978).
The purpose of edge detection is to generate a binary
valued image from a detailed one containing the
boundaries of the scenes or objects within the origi-
nal image.
An edge be enhanced as the boundary between two
regions with relatively distinct gray level properties.
Basically, the idea underlying most edge detection
techniques is the computation of local derivative op-
erator.
The first derivative of an edge modeled in this man-
ner is zero in all regions of constant gray-level,
while the second derivative is zero in all locations
except at the onset and termination of gray-level
transiton. Based on these remarks, it is evident that
the magnitude of the first derivative can be used to
detect the presence of an edge, while the sign of the
second derivative can be used to determine whether
an edge pixel lies on the dark(e.g.background) or
light (object) side of the edge.
The common methods used to calculate the deriva-
tives are the gradient and Laplacian operations.
Several linear window operators were proposed to
simplify the computations and save the time.
According to the following numbering :
al a2 a3
a8 a0 a4 (1)
a7 a6 as
where aj , i= 1,2,.....,8 is the pixel gray-level within
