elastic, highly nonlinear, and therefore, difficult to analyze. For structural description of complex patterns, some Structural deformation caused by discontinuous transformations is an intractable problem in shape analysis and descripglobal, qualitative, structural features are appropriate tion. Structural descriptions depend on the topological strucrather than local, quantitative descriptions based on some ture of the shape, and therefore, they are sensitive to analytical or statistical shape models.
discontinuous transformations which change the topology of
β’ Such discontinuous deformations as caused by stroke the shape. Because of the difficulties, there have been few sysconnections are unique to unconstrained handwritten chartematic studies analyzing and modeling structural deformaacters. The shape description must be robust against and tions caused by discontinuous transformations. In this paper, accommodate such deformations so that various deformed as a first step for overcoming the difficulties, we give a complete patterns can be represented by a small number of classes.
and systematic analysis of structural deformations of curves due to some types of commonly occurring discontinuous trans-
There have been a number of studies conducted over formations, in terms of the curve description based on quasithe past thirty years exploring effective methods for handconvexity/concavity incorporating quantized directional feawritten character recognition [12,13]. In particular, from tures. The transformation laws obtained by the analysis are the viewpoint of shape description and structural feature composed of a small and tractable number of distinct cases. extraction, it has been found that such features as quasi-
The analysis is applied to the automatic construction of class topological features (convexity, concavity, and loop),
from data. In order to show the practical effectivequantized directional features, and singularities (branch ness of the analysis, experimental results are given for unconpoints and crossings) are effective and powerful for handstrained handwritten character recognition using a real and writing recognition. The importance of quasi-topological standard data set. High recognition accuracy is attained for a variety of deformed patterns with a small number of prototypes features in character recognition was first addressed by (typically one class for one character). We also mention some Munson [4, 14]. The idea of Munson was extended to the other applications of the analysis such as the deduction of outermost point method by Yamamoto and Mori [41]. On possible structural descriptions from the class descriptions and the other hand, quantized directional features (four or the examination of the status of conflicts among the class eight directions) [33] have been widely used in character descriptions.