Matching shapes is very important for shape classication
and image retrieval, therefore, shape descriptors play a major rule in Document Image Analyses such as in character
and handwriting recognition, symbol and logo recognition,
or generally speaking shape recognition and matching. In
the literature, we can and several surveys summarizing advances in shape descriptors either in the context of shape
analysis [25] or in the more general context of computer vision and pattern recognition [25, 28]. Dierent taxonomies
of shape descriptors according to different points of view
have been presented trying to make some order in this wide
field. Pavlidis[22], divides the shape descriptors in several
binary classes: external and internal algorithms; scalar and
domain transforms; and information preserving and information non-preserving methods. Mehtre et al.[19], classified shape descriptors as boundary based methods and region based methods. Zhang et al.[28], differentiate between
contour and region based descriptors but they simplify the
classification by only differentiating between structural and
global descriptors. Trier et al.[25] introduced another point
of view distinguishing among features extracted from binary
images and gray-scale images. Another taxonomy divides
them to appearance-based models, where gray or color values of images are directly used to measure similarity, and
feature based methods which use characteristics and descriptors of the target objects. In general, successful description
of a shape contains suffcient information to gather similar and distinguish between different target objects. These
methods can be divided into two categories, the area-based
methods and the boundary-based methods. Simple descriptors, for example perimeter length, curvature, and bending
energy, have been applied widely but proved to be effcient
only as part of a feature set or for eliminating far candidates.
Shapes of the same object can be defined as an equivalence
class under a group of transformations mostly include scale,
translation and small distortions. Shape classifying in such
case, is to belong a given shape to it's equivalence class using
shape similarity measurement. Appearance based method
makes a direct use of gray values within the visible portion of the objects, where feature based methods focus on
the shape geometry. The appearance information is used to
align and find the correspondences of gray scale values to be
compared.
Matching shapes is very important for shape classication
and image retrieval, therefore, shape descriptors play a major rule in Document Image Analyses such as in character
and handwriting recognition, symbol and logo recognition,
or generally speaking shape recognition and matching. In
the literature, we can and several surveys summarizing advances in shape descriptors either in the context of shape
analysis [25] or in the more general context of computer vision and pattern recognition [25, 28]. Dierent taxonomies
of shape descriptors according to different points of view
have been presented trying to make some order in this wide
field. Pavlidis[22], divides the shape descriptors in several
binary classes: external and internal algorithms; scalar and
domain transforms; and information preserving and information non-preserving methods. Mehtre et al.[19], classified shape descriptors as boundary based methods and region based methods. Zhang et al.[28], differentiate between
contour and region based descriptors but they simplify the
classification by only differentiating between structural and
global descriptors. Trier et al.[25] introduced another point
of view distinguishing among features extracted from binary
images and gray-scale images. Another taxonomy divides
them to appearance-based models, where gray or color values of images are directly used to measure similarity, and
feature based methods which use characteristics and descriptors of the target objects. In general, successful description
of a shape contains suffcient information to gather similar and distinguish between different target objects. These
methods can be divided into two categories, the area-based
methods and the boundary-based methods. Simple descriptors, for example perimeter length, curvature, and bending
energy, have been applied widely but proved to be effcient
only as part of a feature set or for eliminating far candidates.
Shapes of the same object can be defined as an equivalence
class under a group of transformations mostly include scale,
translation and small distortions. Shape classifying in such
case, is to belong a given shape to it's equivalence class using
shape similarity measurement. Appearance based method
makes a direct use of gray values within the visible portion of the objects, where feature based methods focus on
the shape geometry. The appearance information is used to
align and find the correspondences of gray scale values to be
compared.
การแปล กรุณารอสักครู่..