Design And Implement Of A System Of Automatic Extraction Of Chinese Character Strokes

Chinese character recognition plays an important role in the development of our state’s information technology. It has broad application prospects, such as document retrieval, office automation, automatic letter classification.On the other hand, Chinese character recognition is also an important component of the scientific research whose development will prompt further achievements in other research areas. With more and more applications of Chinese character recognition, the speed and accuracy of recognition is highly required. Chinese character stroke extraction is a key step in recognition. However, considering the situation that Chinese characters have complex structures, current characters is numerous, the joints, shape of each stroke are various, the relative positions of character components in each character or between two different characters are different, the tilt angle of the same stroke in two characters will change, we are facing a lot of difficulties in automatic stroke extraction for Chinese character.This thesis mainly researches on the high-effective automatic extraction of Chinese character strokes, including character component splitting, shape matching, cross-area detection and stroke segments combination. For the challenge of cross area extraction, we propose an efficient and effective adaptive cross area extraction algorithm which can precisely extract the various cross region of two or more strokes and divide strokes into segments.What’s more, compared with current PBOD algorithm, it reduces the computation time by 95%.In order to improve the performance of stroke segments combination, angle-based combination method is proposed to estimate the probability of two segments combination.In this way, compared with traditional algorithm, we relieve the computation cost, reducing 75% of computation time.On top of these methods, we realise the system of automatic stroke extraction of Chinese characters.Experiments on datasets with different character styles verify the effectiveness and efficiency of our proposed system.