Continuous Chinese Handwriting Recognition System Based On Cloud Platform

As a scientific discipline, human-computer interaction(HCI) has a long research history. With the development of computer technology, intelligent interactive modes have been pursued by more and more researchers, such as graphic interaction, voice interaction and handwritten interaction. These interactions make the communication between human and machine easier. With the rapid raise of touch-screen mobile devices, handwritten recognition has become a more natural interaction because of its robust performance in variance environments.However, most popular handwritten modes are still repeating the same procedure, in which users need to build words, go through ideas, write characters and click candidates. But in this procedure, there are series of problems that users can not continue writing, may break thought and get poor efficiency. Since the progress of technology, the higher accuracy the Recognition Engine has, the more complex structure at the same time. All these limit the recognition on mobile devices. So how to improve the recognition rate in such a background is still a positive challenge in character recognition.To address above issues, we propose a continuous handwriting recognition system based on cloud computing. By combining with the current more popular cloud computing platform, we set up the complex recognition engine based on convolutional neural networks and post-processing algorithms on the cloud server in order to achieve distributed parallel computing that can efficiently solve the shortage of hardware resources. In order to achieve smooth writing, we design a continuous handwritten interaction mode based on cloud. In this mode, we improve the speed of writing and the accuracy of recognition by multi-level rendering and double recognition engine on Android mobile terminal. The synchronization of data between client and server and concurrent user access are kept by building sticky load balancing node in the clouds. We also set up a storage system for user handwritten data, the handwritten data is storage in the database via Network File System(NFS) and the system records user habits in log file. This will play a important role in user data mining, designing of personalized recognition engine and post-processing in the future.At last, we conduct comparative experiments for system. 10 real users write the same document via three writing recognition engines, and their recognition accuracies are compared. The experiment results show that our recognition engine based on the cloud platform gets higher accuracy of 95.70% in first candidate and0.2 second of delay, which satisfied the requirements of ensuring the user’s smooth writing. Next we start up multi-thread on computer to simulate concurrent access to cloud, to test the server load and throughput capacity. The experiment results show that our system owns strong processing power and greater throughput. Finally, we do test on system compatibility through Testin Cloud Test and obtain 98.26%passing rate. The test covers more than 100 different mobile phones which shows great compatibility.