SURVEY ON POLYGONAL APPROXIMATION TECHNIQUES FOR DIGITAL PLANAR CURVES

International Journal of Information Technology, Modeling and Computing (IJITMC) Vol.1, No.2, May 2013
DOI : 10.5121/ijitmc.2013.1206 59
SURVEY ON POLYGONAL APPROXIMATION
TECHNIQUES FOR DIGITAL PLANAR CURVES
Kalaivani Selvakumar and Bimal Kumar Ray
School of Information Technology and Engineering,
VIT University, Vellore, India.
kalaivanis@vit.ac.in
bimalkumarray@vit.ac.in
ABSTRACT
Polygon approximation plays a vital role in abquitious applications like multimedia, geographic and object
recognition. An extensive number of polygonal approximation techniques for digital planar curves have
been proposed over the last decade, but there are no survey papers on recently proposed techniques.
Polygon is a collection of edges and vertices. Objects are represented using edges and vertices or contour
points (ie. polygon). Polygonal approximation is representing the object with less number of dominant
points (less number of edges and vertices). Polygon approximation results in less computational speed and
memory usage. This paper deals with comparative study on polygonal approximation techniques for digital
planar curves with respect to their computation and efficiency.
KEYWORDS
Polygon approximation, contour points, dominant points, digital planar curves.
1. INTRODUCTION
In recent decades, the applications of image processing and computer vision are heuristically
increased. For various types of application like object recognition, industrial dimension
inspection and many other engineering and science fields. Extracting consequential information
from the digital planar curves in many machine vision applications is the important target.
Consequential information means where the point has high curvature and high responsibility for
representing the shape of that object. These points are called as dominant points. The dominant
points are used to represent the shape with less number of contours and retain the shape.
Famous Attneave’s [4] observation is that Consequential information about shape is concentrated
at dominant points of a curve. Those techniques provide a essential data reduction while
preserving important information about object which are used for image matching, pattern
recognition and shape description [5].
Shape representation by polygonal approximation consists of serious vertices. These
representations become very popular for its simplicity, locality, generality and compactness [6].
Polygonal approximation is used in various applications like,

60
• The representation of numerals on car number plates and aircraft by Goyal et al[7].
• Electrooculographic (EOG) biosignal processing by semyonov [8]
• Shape understanding[4]
• By set of feature points achieve image analysis by [9]
• Image matching algorithm.[10,11,12]
The goal of polygonal approximation is to capture the boundary shape with least number of
vertices. In this approach, some techniques achieve optimal polygon approximation but with
maximum error value. [13, 14, 15, 16].
In most of the techniques, the user defined the threshold examine the number and location of
the vertices in the polygon. Some other techniques obtain the vertices directly using dominant
point detection and approximate the polygon. Generally, dominant points are points with high
curvature.
In section 2 the techniques are explained, in section 3 the results are compared with measures.
2. COMPARATIVE STUDY
2.1 Technique 1
Asif Masood[1].This approach approximates the polygon in iterative manner. By using the
freeman’s chaincode, it starts extract the breakpoint from the given contour points.
Fig1:Freeman’s chaincode
4
5
6
3
2
0
7
1

61
Fig 2: After Freeman’s chaincode chromosome breakpoints (in bold)
The figure 1[ref12] shows the freeman’s chaincode shows all possible direction of the points.
Consider figure 2, the chromosomes mark with start indicates the start direction for freeman’s
chaincode.
C = {c1,c2,…………………,ci-1,cn}
P= {(x1,y1),(x2,y2),……(xi-1,yi-1),(xn,yn)}
i = { 1,2,3,……………i-1,n}
where C is a collection of freeman’s chaincode value (from 0 to 7) and n represents the number of
contour points. The P is a collection of contour point’s x and y co-ordinates for n number of
contour points.
Table 1. Freeman’s Chaincode
The associated error value (AEV) is calculated using those extracted breakpoints. The
perpendicular (squared) distance between point Pk to the straight line Pi and Pj is called associated
error value (AEV) .Take sequence of points as Pi, Pj and Pk and calculate the AEV as follows,
Table 2.Calculation of AEV
If Ci ≠ Ci+1 Then
Pi is a breakpoint.
Else
Pi is not a breakpoint.
End
1. Join Pi and Pj as a straight line
2. Draw perpendicular line from Pk point which joins the
line Pi and Pj
3. Calculate perpendicular squared distance where
Pi ≡ (xi ,yi) , Pj≡ (xj ,yj) and Pk≡ (xk ,yk)

62
22
2
)()(
)))(())(((
jiji
ijikijik
yyxx
xxyyyyxx
−+−
−−−−−
The AEV is calculated for each break point and stored. The algorithm compares the AEV
sequence and identified the least AEV’s break point to be deleted. This iteration process deletes
all redundant breakpoint. This technique produces expected quality result but still suboptimal, to
address this stabilization process is introduced. Stabilization algorithm relocates the break points
by comparing its integrated square error (ISE). Integrated square error (ISE) is calculated as
follows:
∑=
=
n
i
ieISE
1
where ei is the squared distance of ith point P of the approximated polygon. This process
compares ISE values of its predecessor and successor and relocates the break point where its
shape doesn’t make more change. It reduces the total distortion (error) which is caused by its
neighboring vertices. By this stabilization process the dominant point (DP) moves to optimal
location and produces optimal approximated polygon. This approximation technique results at
any compression ratio or required error or at any finite approximated measurement.
Table 3: Masood’s polygonal approximation algorithm through stabilization process
Number of contour points is N
Apply chaincode using Table 1
Calculate AEV for all Contours using Table 2
While(termination condition) and store
Drop DP with least AEV
Mark flag for neighboring DPs
Apply stabilization procedure
do
List = collection of unstable DPs
If empty (List)
Return ( )
End if
For each unstable DP
SP=set stable position
If SP ≠ current position of DP
Move DP to SP

63
This approach iteratively deletes the redundant point and stabilization process, stabilizes the points in right
location with lass error value and produce high quality polygon approximation.
2.2 Technique 2
Carmona [2] proposed a new breakpoint suppression algorithm which suppresses the redundant
point and detects the dominant point effectively. This approach first applies freeman’s chaincode
and extracted initial breakpoints using [ref2]. Then perpendicular distance is calculated using
[refence].Select any breakpoint from the set of breakpoints which is extracted using chaincode.
Assume a threshold value dt which is less than one. Now compare the perpendicular distance d
sequence with threshold value dt if this d value is less than dt, remain that points which is liable to
that distance. This iteration follows until the final point is equal to initial points.
Table 4: Carmano’s polygonal approximation algorithm through break point suppression process
Calculate perpendicular distance d for all Contours using Table 2
Select initial point Pini1
Set threshold value dt value < 1 (e.g. 0.1)
If d < = dt
Then eliminate Pj
Pj = Pj+1 and Pj+1 = Pj+2
Else
Pj-1= Pj, Pj= Pj+1 and Pj+1= Pj+2
End if
Until Pj = Pini1
Increase dt
Repeat end
Mark flag for neighboring DPs
End if
Get unmark DP
End for
While (no unstable points)
List =get repositioned DPs
Recalculate and update the AEV for List

64
The termination condition is based on the ratio of length associated with each deleted contour
point Pi and the maximum error obtained for each contour point Pj. A simple and efficient
approach for polygonal approximation is discussed. This algorithm produces a polygon
approximation with any number of dominant points as per the threshold value.
2.3 Technique 3:
Tanvir Parvez [3] proposed a new adaptive optimization algorithm for polygonal approximation.
This technique is robust to noise and effective in polygon approximation. It concentrates on the
point where its curvature falls high. With those points, the optimization procedure is applied to
get the best-fitting polygon approximation. This algorithm follows the points where the level of
information is high. Suppression technique is used to address the noise and produce good results.
The technique works well with Arabic letters and MPEG CE7 Shape-1Part B database. The
algorithm first performs binarization and smoothing using Otsu (1979) and statistical average
smoothing is applied (Mahmoud, 1994). With the smoothened image, the contours are extracted
using the algorithm in (Pavlidis, 1982).
The selection of breakpoints is extracted using freeman’s chaincode. The set of cut-points are
extracted using Constrained Collinear points (CSS) and remove redundant dominant point. To
detect the final set of dominant points, the initial value goes on increasing and will reduces the set
of cut-points. Finally, optimization procedure obtained the least candidate sets of points of the
polygonal approximation.
Table 5: Tanvir Parvez’s adaptive optimization algorithm for polygonal approximation
Smooth the binary image and extract the contour.
Apply Freeman’s chaincode using table 1
Calculate the left support region and right support region for refined break points
Calculate the centroid for all break points.
Sort the dominant points based on strength first, then based on distance from the
centroid to select weakest point first.
For each dominant point from the sorted list , suppress the point if all three conditions
are satisﬁed.
Set threshold value as 0.5, if the perpendicular distance is less than threshold.
- Triangle formed by three consecutive points must be acute triangle.
- For each unsuppressed point Pl other than Pi, Pj and Pk, the minimum
distance from Pl to the line-segment joining Pi and Pk is greater than dcol.
- Update the strength of the other candidate dominant points and sort them.
Repeat until we obtain optimize efficiency measure elect weakest point first.

65
3. RESULTS AND DISCUSSION
The measures taken for comparison are nd, CR, ISE, WE.
• Compression ratio (CR)
CR = n/nd
where n is the number of points in the contour and nd is then number of points of the polygonal
approximation.
• The Integral Square Error (ISE) is defined as
where ei is the squared distance of ith point P of the approximated polygon.
• Weighted sum of square error,
WE = ISE/CR
The modiﬁed version of WE measure is used by Marji and Siy (2004) as
WEx = ISE/CRx
where x is used to control the contribution of the denominator to the overall result in order to
reduce the imbalance between the two terms.
The experimental results discuss the approximated polygon for the standard shapes like
chromosomes,
Table 6. Comparison of Masood’s, Carmona’s and Tanvir’s method with standard shapes like
chromosomes, leaf, semicircle and infinity.
Shapes Method nd CR ISE WE WE2 WE3
Chromosome (n =
60)
Masood-2008 15 4 3.88 0.97 0.243 0.061
Carmona-Poyato-2010 15 4 4.27 1.07 0.267 0.067
Tanvir-2010 10 6 14.34 2.39 0.398 0.066
Leaf (n = 120) Masood-2008 23 5.22 9.46 1.81 0.347 0.067
Carmona-Poyato -2010 23 5.22 10.68 2.05 0.391 0.075
Tanvir -2010 21 5.71 13.82 2.42 0.423 0.074
Semicircles (n =
102)
Masood-2008 26 3.92 4.05 1.03 0.263 0.067
Carmona-Poyato -2010 26 3.92 4.91 1.25 0.319 0.082
Tanvir -2010 17 6 19.02 3.17 0.528 0.088
Inﬁnity (n = 45) Masood-2008 11 4.09 2.9 0.71 0.173 0.042
Carmona-Poyato -2010 10 4.5 5.29 1.18 0.261 0.058
Tanvir -2010 9 5 7.35 1.47 0.294 0.059
∑=
=
n
i
ieISE
1

66

67
Fig 3.Graph shows the performance f three methods for the standard shape.
leaf, semicircle and infinity. For the comparison, the standard datasets like chromosomes, leaf,
semicircle and infinity has been used. For the chromosome shape, Tanvir’s results least number
of dominant points. The ISE and WE values are high while comparing with other two techniques.
For the leaf shape, Tanvir’s results least number of dominant points. The measure WE3 of
Masood’s results is better while comparing with other two techniques. For the semicircle shape,
Tanvir’s results least number of dominant points. Masood’s result lowest WE3 while comparing
with other two techniques. For the infinity shape, Tanvir’s results least number of dominant
points. Masood’s results least WE3 while comparing wither two techniques.
Table 7. The results comparison of three techniques with standard datasets

68
4. CONCLUSION
In this paper, the surveys of the three techniques are clearly illustrated and the results are
compared with suitable measures. This paper fully focuses on the ideas which are explored in
each technique are gathered and are elaborately explained with suitable datasets. With the help of
graph the measures for each shape are calculated and are compared with other techniques.

69
REFERENCES
[1] Carmona-Poyato, A., Madrid-Cuevas, F.J., Medina-Carnicer, R., Muñoz-Salinas, R.,Polygonal
approximation of digital planar curves through break point suppression. Pattern Recognition 43
(2010), 14–25.
[2] A. Masood, Dominant point deletion by reverse polygonization of digital curves, Image and Vision
Computing 26 (2008) 702–715.
[3] Mohammad Tanvir Parvez , Sabri A. Mahmoud, Polygonal approximation of digital planar curves
through adaptive optimizations, Pattern Recognition 31 (2010)1997-2005.
[4] F. Attneave, Some informational aspects of visual perception, Psychological Review 61 (1954) 189–
193.
[5] S. Hsin-Teng, H. Wu-Chih, A rotationally invariant two-phase scheme for corner detection, Pattern
Recognition 29 (1996) 819–828.
[6] A. Melkman, J. O'Rourke, On polygonal chain approximation, in: G.T. Toussaint (Ed.),
Computational Morphology, North-Holland, Amsterdam, 1998, pp. 87–95.
[7] S. Goyal, M.P. Kumar, C.V. Jawahar, P.J. Narayanam, Polygon approximation of closed curves
across multiple views, in: Indian Conference on Vision, Graphics and Image Processing, 2002.
[8] P.A. Semyonov, Optimized unjoined linear approximation and its application for Eog-biosignal
processing, in: 12th IEEE International Conference on Engineering.
[9] R. Neumann, G. Teisseron, Extraction of dominant points by estimation of the contour fluctuations,
Pattern Recognition 35 (2002) 1447–1462.
[10] X. Hu, N. Ahuja, Matching point features with ordered geometric, rigidity, and disparity constraints,
IEEE Transactions on Pattern Analysis and Machine Intelligence 16 (1994) 1041–1049
[11] I.K. Sethi, R. Jain, Finding trajectories of feature points in a monocular image sequence, IEEE
Transactions on Pattern Analysis and Machine Intelligence 9(1987) 56–73.
[12] P.C. Yuen, Dominant points matching algorithm, Electronic Letters 29 (1993)2023–2024.
[13] J.C. Perez, E. Vidal, Optimum polygonal approximation of digitized curves, Pattern Recognition
Letters 15 (1994) 743–750.
[14] A. Pikaz, I. Dinstein, Optimal polygonal approximation of digital curves, Pattern Recognition 28
(1995) 373–379.
[15] M. Salotti, An efficient algorithm for the optimal polygonal approximation of digitized curves,
Pattern Recognition Letters 22 (2001) 215–221.
[16] M. Salotti, Optimal polygonal approximation of digitized curves using the sum of square deviations
criterion, Pattern Recognition 35 (2002) 435–443 (Marc Salotti).
[17] MajedMarji, Pepe Siy, Polygonal representation of digital planar curves through dominant point
detection—a nonparametric algorithm, pattern recognition 37,2004,2113-2130.
[18] Pavlidis, T., 1982. Algorithms for Graphics and Image Processing. Computer Science Press,
Rockville, MD.

SURVEY ON POLYGONAL APPROXIMATION TECHNIQUES FOR DIGITAL PLANAR CURVES

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to SURVEY ON POLYGONAL APPROXIMATION TECHNIQUES FOR DIGITAL PLANAR CURVES (20)

Recently uploaded (20)

SURVEY ON POLYGONAL APPROXIMATION TECHNIQUES FOR DIGITAL PLANAR CURVES