Image Processing Reference
In-Depth Information
in R2. However, we have dropped R3 and R4, since they impose very tight
restrictions on S to be recognized as a DSS. Such a policy has been done in
order to successfully extract the ADSS, and some of the advantages are as
follows:
• avoiding tight DSS constraints, especially while representing the gross
pattern of a real-world image with digital aberrations/imperfections;
• enabling extraction of ADSS from a curve segment, thereby straighten-
ing a part of it when the concerned part is not exactly digitally straight;
• reducing the number of extracted segments, thereby decreasing storage
requirements and run-time in subsequent applications;
• reducing the CPU time of ADSS extraction; and
• usage of integer operations only (e.g., to compute ⌊(p + 3)/4⌋, 3 is added
with p, followed by two successive right shifts).
Since the chain code of a curve segment is taken in a one-dimensional list,
S, the ADSS may be characterized by the following sets of parameters:
• Orientations parameters:
n
(non-singular element),
s
(singular element),
l (length of leftmost run of
n
), and r (length of rightmost run of
n
).
They play decisive roles on the orientation (and the digital composition,
thereof) of the concerned ADSS. For example, in Fig. 4.4, the curve S
1
has
n
= 0,
s
= 1, and chain code 0
5
10
6
10
6
10
5
10
5
10
6
having l = 5 and
r = 6.
• Run-length interval parameters: p and q, where [p,q] is the range of
possible lengths (excepting l and r) of
n
in S that determines the level
of approximation of the ADSS, subject to the following two conditions:
(c1) q−p 6 d = round(p/2).
(4.7)
(c2)
(l−p),(r−p) 6 e = round(p/2).
(4.8)
While implementing DETECT-ADSS, we strictly adhere to R1, as it is
directly related to the overall straightness of S. However, we have modified the
stricture in R2 by considering that the run lengths of
n
can vary by more than
unity, depending on the minimum run length of
n
. The rationale of modifying
R2 to Condition c1 is that, while approximating the extracted line segments
from S, an allowance of approximation (d) specified by c1 is permitted. Given
a value of p, the amount d by which q is in excess of p indicates the deviation
of the ADSS from the actual/real line, since ideally (for a DSS) q can exceed
from p by at most unity (the significance of d in characterizing an ADSS is
detailed out in [13]).
Apart from d, the other parameter, namely e, is incorporated in c2, which,
along with c1, ensures that the extracted ADSS is not badly approximated
