Segmentation and Recognition of Handwritten Chinese Day String

Segmentation and recognition of off-line handwritten Chinese character string is a difficult task in the research field of character recognition. A standard way for character string recognition is to segment a string into isolate character, then compos their recognition results into words or strings. The purpose of segmentation is to reduce the pattern classes which are to be sent to the recognition engines. However, recognition failure caused by segmentation line missing, non character patterns and unreliable recognition scores. To recognize the Chinese day strings on check images, a rule based method is proposed. It recognizes date strings by combining a holistic method and a segmentation-recognition based method. The holistic method recognizes the whole string as a single character without segmentation. The segmentation-recognition based method first finds as much candidate segmentation lines as possible by projection and structure analysis. Then, it reduces segmentation lines by a predicted string length. Finally, the best recognition result is selected by recognition scores. Experiments have been done on 5569 real life check images collected from Chinese bank. The experiment results demonstrate the efficiency of the proposed method. The string recognition rate has achieved 93.3% on the 1932 test strings.