Few researches have focused on translation-, scale-, and rotation-invariant recognition of handwritten Chinese characters since writing variation is large and invariant features are hard to find. In this study, an adequate normalization process is first used to normalize characters such that they are invariant to translation and scale; then, five rotation-invariant features are extracted for invariant recognition. We propose a feature-extraction approach which especially considers the skeleton-distortion problem to effectively extract the desired invariant features. The first four invariant features are used for preclassification to reduce the matching time. The last feature, ring data, is used to construct ring-data vectors and weighted ring-data matrices to individually characterize character samples and characters for invariant recognition. A character set was constructed from 200 handwritten Chinese characters with several different samples of each character in arbitrary orientation. Several experiments were conducted using the character set to evaluate the performance of the proposed preclassification and ring-data matching algorithms. The experimental results show that the proposed approaches work well for invariant handwritten Chinese character recognition and are superior to the moment-invariant matching approach.
