Learning network and structural based character recognition

Abstract

The advent of Tablet PC suggests the important future role of pen-based handwriting recognition in the new era of computing technology. Recognition of handwritten Chinese characters has been extremely challenging because of the large number of available characters, large character shape variations, various handwriting styles and unpredictable dynamic information in online handwriting. The popular statistical pattern matching (SPM) approaches and online structural approaches in general both have their own pros and cons, and they are mutually complementary. SPM approaches usually ensure achievement of reliable baseline recognition accuracy but it is inefficient in recognizing simple Chinese characters and in distinguishing some similar characters. Online structural approaches are more flexible since human knowledge is frequently used in the form of heuristic rules, but it is computationally expensive when an input character has too many structural primitives.