Research Of Screen Soft Keyboard Automatic Identification Technology

Along with the rapid development of Internet technology,network entertainment,Internet transactions and the network bank interactive network platform are widely applied.In order to strengthen the safety,users often need to login in the platform loginning interface with screen soft keyboard.Because of using the soft keyboard so complicatedly and fussily,it s in urgent need of a tool that can automatically identify screen soft keyboard to help users increase efficiency.And at present, there s no software that can automatically identify characters in the specific soft keyboard of a web page on the market yet.Based on this,we decide to use the knowledge what we have learned, innovate and improve the advanced image processing and pattern recognition algorithm,develop a character recognition system with application value--screen soft keyboard automatic identification system.This paper firstly researches on searching the keyboard region of on the internet banking page.By comparing with several image matching methods more commonly used at present and combinating of image pyramid thought, we propose a template matching strategy algorithm based on block and pyramid.This algorithm inherits the accuracy of template matching algorithm.And at the same time it can drastically reduce the amount of calculation,while assure the accuracy of the template matching.Experimental results show that by using this method,we can quickly and accurately finish the process of template matching and find soft keyboard in colorful web pages.Secondly at the part of segmentation virtual keys in the keyboard area,we make the experimental analysis on existing image edge detection methods and edge extraction methods.Finally we put forward a character segmentation method using Flood Fill method and elliptic fitting way.Experimental results show that this method can effectively segment each character accurately,and make the good foundation for the subsequent accurate feature extraction and character recognition.And then at the part of character recognition, feature extraction and random decision tree are discussed.We find out 29 features appropriate for the system,and design a three level cascade random training tree classifier.This classifier make full use of its own characteristics.By the first two levels broad classifying, it can narrow the third level s identification range and make precise classification by using random decision tree.Experimental results show that this method improves the character recognition training efficiency and recognition rate.Finally based on the algorithms researched above in this paper,we design and realize the screen soft keyboard automatic identification system.The article discuss on the system overall structure, each module design and implementation,and the performance of system in detail.We also randomly selected various types of internet bank landing interface,and do the experiment with our system.We evaluate the system from real-time,stability, expandability and other performance indexes.A lot of experimental results show that system satisfies the predetermined performance indicators from the accuracy of web soft keyboard identification,the precision of character segmentation,character recognition accuracy and other performance indexes.