The Application Of Improved Image Registration Method In Chinese Character Recognition

Chinese characters, in the form of which the Chinese civilization is recorded, have thousands of years of history. However, in this information era, it is difficult to handle huge amounts of information for information explosion. Therefore, how to make the computer obtain the information of Chinese character efficiently to save human resources becomes a key issue. And the best solution is Chinese character recognition. The current researches of Chinese character recognition based on OCR(Optical Character Recognition) system have achieved good result. But it is still inadequate for the complex image which has movement, rotation, etc. So we propose a Chinese character recognition which was based on image matching algorithm and experimental results verified the feasibility of this algorithm. This algorithm provides a new research idea for the Chinese character recognition.The work is as follow:Firstly, in this paper we analyzed the characteristics of commonly used Chinese characters, Chinese character recognition algorithm and image matching algorithm. And we emphasized on a kind of Chinese character recognition method based on the shortest European distance.Secondly, in order to introduce the image matching algorithm into the Chinese character recognition, we improved the image matching algorithm. However, there were a lot of false matching point pairs after point matching which had great influence on the results of recognition. Thus we put forward a fault elimination matching method based on distance constraint. This method substantially reduced the false matching between different characters.Thirdly, in order to restraint the decreasing recognition rate for the the complex image recognition, we proposed a two stage Chinese character recognition algorithm based on improved image matching algorithm. The algorithm firstly extracts the external characteristic of the Chinese character image for coarse classification, then uses SURF algorithm to describe the image feature for precise classification. The improved matching algorithm was used to match the SURF feature points, and finally the recognition process was completed by calculating the number of matching points. The experiment results showed that this algorithm was feasible and achieved good recognition.