Chinese Text Map To Deal With Handwritten Chinese Character Recognition

This dissertation is devoted to the process of HCTG(Handwritten Chinese Text Graph) and off-line HCCR(Handwritten Chinese Character Recognition). The emphases include binarization of colorful HCTG, HCTG noise elimination, lines extraction from HCTG, words extraction from text line, HCCR based on feature extraction and classification, strokes extraction and HCCR based on structural matching.The great part of our work is focused on the process of HCTG. Chapter 1 introduced the un-uniformity of the brightness distribution and the color distribution of the colorful graphs firstly. And then according to the idea of binarization part by part, we implemented the binarization of colorful HCTG by applying the binarization formula and threshold adjustment. In chapter 2, we offered some methods for HCTG noise elimination including the method of CC(connected component) area, the method of strokes expansion and the method of snow thaw etc. With the method of CC area and the method of snow thaw, we cleared the little spots, holes and isolated lines from text graph. And we cleared the isolated points, protruding points and sunken points too. For lines in HCTG, we introduced several methods to detect them. The projection curve analysis was used for detection of direction-known lines. And Hough transform method is the common detection method. In addition we also discussed methods like "sewing work" and "excitated projection" etc. In chapter 4, we analyzed the projection curve created with feature CBSN (Consecutive Blank Segments Number), and found the possibility of only one wave crest to each line and one trough between two lines and its stability. Based on this phenomenon, a method for lines extraction from HCTG was introduced.As to words extraction from text line, chapter 5 introduced a method for words extraction based on CC analysis, and then introduced a method for words extraction based on inertial process.In HCCR part, we introduced HCCR based on feature extraction and classification, strokes extraction and HCCR based on structural matching. Chapter 6 introduced the uniformization of the single character HCTG. In chapter 7, we firstly introduced several features used to make up the vector of statistical feature for Chinese character. Then with the feature selected, we can experiment the HCCR based on feature extraction and classification. With the strokes extracted by the methods introduced in chapter 8,we discussed the problems of HCCR based on structural matching in chapter 9.In summary, we have found some good schemes for the process of HCTG and HCCR, such as binarization of colorful HCTG, HCTG noise elimination and lines extraction from HCTG But some still need more experiments to obtain high stable algorithms, such as for extraction of lines in HCTG, words extraction from text line, the pre-process for Chinese character, HCCR based on feature extraction and classification and strokes extraction. And some are still under theory construction, which should be proved by experiments. These include HCCR based on structural matching, post-process of HCCR etc.