Research On Offline Handwritten Chinese Character Segmentation

Offline handwritten Chinese character segmentation is the foundation ofoffline handwritten Chinese character recognition, the accuracy of offlinehandwritten Chinese character segmentation directly affects the rate ofrecognition. Researches on offline handwritten Chinese character segmentationhave very vital significance for improving the overall system performance ofoffline handwritten Chinese character recognition. This thesis takes offlinehandwritten Chinese character as the research object, makes studies of problemsrelated Chinese characters projection segmentation, and makes correspondingimprove at last.The main work and innovation as follows:1) In the process of writing Chinese characters, it may exit phenomenonsthat characters adhesion and characters overlapping, which could reduce theaccuracy of offline handwritten Chinese character segmentation. In order tosolve the problem, the thesis puts forward Clearance Project SegmentationAlgorithm Based on multi-threshold and Multi-stage Cut. This algorithm usesmultiple thresholds to do segmentation several times, and then selects thesegmentation scheme which has the best segmentation result. Besides, thealgorithm divides segmentation into several stages, deals with the estrangedcharacters and close characters in different phases. Experiments show that,compared with traditional algorithm of projection segmentation of Chinesecharacters, this algorithm has improved accuracy of segmentation.2) Base on Clearance Project Segmentation Algorithm Based on Multiplethresholds and Multi-stage Cut, this article puts forward a algorithm namedVertical Projection Characters Segmentation Based on Minimum Threshold andCurve Fitting, considered factors which influence the accuracy of thesegmentation, joins identification of punctuations, revision of nonstandardstrokes, and length statistics of segmentation pieces. Because multi-thresholdsegmentation is very verbose and subjective, the thesis uses minimum threshold instead of multi-threshold and make merge strategy in the final stage.Considering uncertainty of looking for segmentation points, this algorithm usescurve fitting methods to look for cut points instead of threshold method.Comparing with the general projection segmentation algorithm, this algorithmhas higher segmentation efficiency and wider applicability.