Personalized Handwritten Chinese Character Recognition System Based On Cloud Computing

With the rapid development of computer technology and mobile communicationtechnology, the mobile devices which has a touch screen, such as PDA, smart phone andtables, are becoming more and more popular. Handwritten input methods as one of the mostnatural input way, along with the development of touch screen and mobile devices, are beingused as the major input method by more and more people.SCUT gPen is the first online handwritten Chinese character recognition input methodapp combined with cloud computing. In the case of the user’s permission, the user’shandwriting trajectory is sent to the cloud module for recognition. With the help of the cloudrecognition module, it can improve the recognition rate, and store massive users’ handwritingsample data into the server side’s database. All these data are written in an unconstrained way.Due to the limitation when training a Chinese character recognition classifier, and greatdifferent writing style among each user, general Chinese character recognition classifier couldnot satisfy all the users’ need. If we can collect massive user’s sample writing data, we canuse incremental leaning method to perform better recognition rate. But all these studies havenot been validated in real usage scenarios yet, on the other hand, how to use cloud computingtechnology to storage the massive handwriting sample data and perform data mining has littlein-depth study.In this paper, we are going to study these two problems. Combined with the cloudcomputing platform, we proposed and implemented a overall personalized handwrittenChinese character recognition framework, which is able to recognize Chinese character in theserver side, store massive handwriting sample data, perform data mining and the writing styleadaptive training and testing, in a semi-automate way. The main task include:(1) Under the permission of the user, the user’s writing trajectory is sent to the cloudmodule for recognition through the gPen client, and save to the server side’s database. Usingthe recognition module in the server side, this is the first report in the online Chinesecharacter recognition study, one can collect20million users, totally150million writingsample data in less than one year, and increase about2million writing sample data daily, allthese data are written in an unconstrained way. In this way, we can collect massive handwriting sample data for follow-up study of handwriting character recognition.(2) In the cloud platform, we designed and implemented the massive handwriting datamodule to access and store the sample data, so we can quickly and easily retrieval and miningthe massive data set. Using the massive handwriting data set, we perform different type ofdata mining task, and found out some interesting statistics on these users’ writing habits,writing preferences and geographical distribution, all in an unconstrained way. This stepprovide massive data and useful experience for the ongoing in-depth study to mining thesample data.(3) Using the statistics method develop earlier, we can select appropriate user forincremental learning, in a semi-automated way. Using this method, we can verified that, wecan collect massive user’s handwriting sample data and then using the incremental learningmethod to increase certain user’s recognition rate. The experiment result shows that theincremental learning theory can be applied to a real life scenario.(4) Optimize the accessing and processing performance for stand-alone machine, analysisthe bottleneck of the platform and found out the solution, making the stand-alone machine canbe working in all day long. The stand-alone server can response1.8million handwritingrecognition request every day. While taking advantage of high availability and high scalabilityadvantage of cloud computing, with the growing number of users and computing needs, wecan expand the cloud computing platform in a smooth way and without affect the user’snormal request.Combined with cloud computing technology and the personalized handwritten Chinesecharacter recognition framework proposed in this paper, we provide useful data andexperience for more subsequent in-depth study of the handwritten Chinese characterrecognition.