cations were made to this method to reduce the memory
requirements and reduce the processing time. These
modifications are done by using the developed coded
edge images.
The thinning algorithm is followed by a linking charac-
teristics of the implemented by analyzing the character-
istics if the pixel neighborhoods, this analyses depends
on the edge code.
The next step is removing the separate points by using
the point detection technique discussed in section 2.1.
The point is removed if the result of convolving the im-
age with the mask given in (Figure 1) is equal to (8).
The resultant edge image obtained above, which con-
tains a thin linked coded edges, undergoes a tracing pro-
cess, in which the edge elements are traced according to
their orientation to form a straight line.
For each starting point ((a point which has no predeces-
sor)) the tracing is carried out by seeking for the succes-
sor of this point, the seeking process is repeated from
the current position until a point with no successor is
reached. This point is recorded as an end point.
The obtained lines are identified by their two end points.
Finally short lines are filtered out to reduce the possible
corresponding ambiguity.
4. Straight Lines Matching
The matching algorithm finds corresponding pairs of
lines from a pair of images, based on a match finction.
This function combines several desripitive parameters-
such as , length, orientation, light edge which may be
defined as the average of lightest 1046 of the gradient
magnitude around the edge, (Mclutosh,1988), and end
points coordinates of each line which are weighted ac-
cording to their relative importance. Matched lines are
those which have the largest match function values and
correspond each other in both directions.
The match finction measures the similarity of the par-
ameters attached to each line and is calculated for the ith
parameter as followes:
Mr= >, Vi Wi (19)
i
1z 1,2. NER 1.25:M
where N is the number of lines in image L and M is the
number of lines in image R,vi is the similarity value of
the ith parameter and wi is the weight corresponding to
the ith parameter, the sum of all the weights is (1)
RS wi =1 (20)
1
Asimilarity measurement for all parameters ((but the an-
gle and disparity parameters)) is defined as the ratio of
the minimum of the parameter values for the two lines
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
divided by the maximum to the two parameter. The sim-
ilarity value v is defined as :
Min [ param; (line .), param; (line r)]
Vi= (21)
Max[ param; (line ), param; (line r)]
where OS vi S 1
An absolute difference in values does not work as well
due to the difference of measurement. The angle and
disparity parameters must be compared in a slightly dif-
ferent manner because only the absolute difference be-
tween the two parameter values is relevant. This differ-
ence is compared to the expected maximum difference.
After similarity values Vj are calculated, a two similarity
measures of each line from image (L) to each line from
image (R). Each element in the array is the sum of the
similarity values for each parameter multiplied by its
corresponding weight value.
The similarity parameters that were used here in the im-
plemented matching algorithm are : length, midpoint co-
ordinates and orientation. The result are shown in Figure
(8).
CONCLUSION
In this paper, we have described automatic system for
stereo image matching, a system that uses edge based
method.
The prime objective of this paper was to test the feasi-
bility of using edges to match — areal stereo pair , we
developed an algorithm for extracting straight lines
which starts with edge detection followed by edge link-
ing, edge thinning, isolated points removing and points
traeing to form the straight fines. Match function is cal-
culated using line length orientation & end points coor-
dinates to determine the similarity measure between the
lines.
REFERENCES
- Ballord D.and Brown C.,1982. Computer vision, prin-
tice- Hall Inc.
- Gonzalez R. and Wiulz P., 1987. Digital image pro-
cessing, Addision-Wesley Pup.Con.Inc.
- Gireenfeld J.S. and Schenk A.F., 1989. Experiments
with edge based stereo Matching, Photag. Eng. and Re-
mate Sensing Vol.55 PP.171-180.
- Mcintosh J.H. and mAtch K.M., 1988. Matchinstraight
lines, computer vision, graph and image processing
Vol.40.PP.386-408.
- Nevatia R. and Babu K., 1980. Linears Feature Extrac-
tion and description,Conp. graph and image proc.
Vol.13 PP.257-269.
- Pratt W.,1978 Degital image processing John wiely
Sns Inc.
- Sahoo P. :
vey of thres
and image p
-Toth C. an
outomaticre
Figure
a) Usu
b) Usi
c) Usu
d) Usii
