ABSTRACT
Polygonal approximation is a process to approximate an arbitrary curve with a sequence of straight lines which retain the overall shape of the original curve. Such a process is useful in shape representation [23, 109, 140, 175, 236] or in data reduction [168, 202]. Attneave [6] suggested that corners or high curvature points of a curve provide important information during the recognition process in the human visual system. When people look at an image, they first capture such information. Such points (i.e., feature points) of a planar curve capture crucial shape information and can potentially be detected consistently. Based on this observation, one can argue that a planar curve can be represented by another planar curve and still be recognized to have the same shape by a human observer if the feature points remain invariant [117]. Thus the polygonal approximation can be formulated as a feature point detection problem. The approximated curve is one with the detected feature points connected by straight line segments [82, 229]. In the logo recognition system, feature point detection is one of the important modules. It is applied to the contours of logos and extracts a set of consistent feature points of a contour. Data reduction can be achieved by using line segments to connect these points to represent shape efficiently. Computation time can thus be saved. In addition, consistent feature points can be used to aid the normalization process which affects the recognition greatly.
