A New Affine Invariant Descriptor for Shape Recognition
Wei Wang*
College of Optoelectronic Science and Engineering,
National University of Defense Technology
Changsha, China
e-mail: wangwei8610@nudt.edu.cn
Lingjun Zhao
College of Electronic Science and Engineering,
National University of Defense Technology
Changsha, China
e-mail: nudtzlj@163.com
Xingwei Yan
College Nine
National University of Defense Technology
Changsha, China
e-mail: yanxingwei@nudt.edu.cn
Jianhua Shi
College of Optoelectronic Science and Engineering,
National University of Defense Technology
Changsha, China
e-mail: gexin7651@sina.com
Abstract—Shape is considered to be one of the most promising
tools to represent and recognize an object. In this paper, an
effective and rigorous shape descriptor is developed for shape
recognition under affine transformations. To reflect the gray
information of the shape, two extended centroids with different
orders are adopted to form the vector which is treated as the x-
axis to establish a new positively oriented orthonormal frame.
Then, the descriptor representing the distribution of all points
within the shape region along the y-axis is captured to achieve
shape recognition. In addition to its stable affine invariance
and robustness to noise, the descriptor has low computational
complexity. The proposed descriptor has been implemented on
a famous database and gives encouraging results while it is
compared to other two state-of-the-art algorithms including
GC and EC-ARC.
Keywords-extended centroids; shape projection; affine
invariant descriptor; shape recognition
I. INTRODUCTION
Pattern recognition has been widely used in a number of
applications such as fingerprint identification [1], sence
matching [2], robot detection [3] and so on. As shape is one
of basic features used to describe image content, shape based
object recognition is an important topic in the area of pattern
recognition. However, shape representation and description
is a difficult task especially when the shape is perturbed by
distortions and noise. In recent years, researchers have
proposed a number of shape description methods which can
by generally classified into two classes: contour based
methods and region based methods.
Contour based shape description only exploits shape
boundary information and achieves good performance [4].
Most contour descriptors have been proposed based on point
sets, curvatures, curves, and so on. In a great deal of
algorithms, the contour is represented as a point set. In [5], a
descriptor, shape context (SC), is proposed to describe the
coarse distribution of the rest points of the shape with respect
to a given point. SC is then enhanced in [6] by using an
ordered Bag of Feature model [7]. Moment [8] is also one
of important point based shape description. The invariance of
descriptors is highly dependent on the assumption that the
support point set is invariant. However, the assumption is not
always valid. In [9], Wang et al. proposed a method to get
invariant support point sets under affine transformations. As
the support points are from the contour, they are sensitive to
contour extraction. The curvature scale space (CSS) [10] is
one of the most well-known curvature based shape
descriptions. In CSS, the position of inflection points are
tracked in the boundaries which are filtered by low-pass
Gaussian filters of variable widths and a curvature scale
space contour map consisting of inflection points is resulted
from the smoothing process. The peaks of individual
branches in the curvature scale space contour map are
detected as feature points to represent the shape. Except for
the feature points, curve segments are always taken as
primitives for shape representation while the contour is
decomposed based on curvatures [11], polygonal
approximation [12], B-spline [13], [14], and so on. Similar to
curvature based methods, curve based method also proves to
be complex and expensive. Generally speaking, as only a
small part of shape information is captured, the contour
based shape descriptors are sensitive to noise, and what is
more, it is incapable of handling the shapes with same
contours and different context.
Compared to contour based shape descriptors, region
based shape descriptors are able to represent the shape more
effectively as all pixels in the shape region are taken into
account. Shape decomposition, shape structural and relative
spatial relationship are well studied region based shape
representation. Similar to contour based shape descriptors,
these shape descriptors can’t cope well with the shapes with
same contours and different context as they don’t take the
gray value of pixels into account. For example, Wang et al.
[15] designed a region based shape descriptor to describe the
shape projection distribution (SPD) along the vectors from
the centroid to each sampled contour point. Hasegawa and
Tabbone [16] combine the Radon transform, the amplitude
1693
2017 3rd IEEE International Conference on Computer and Communications
978-1-5090-6351-2/17/$31.00 ©2017 IEEE