| There are millions of airline coupons need to be processed by the companies everyday. In China, airline coupons must be stored in a warehouse more than 5 years. The airline coupon information, such as ticket-number, fare, seat-class, must be stored in a database. It needs a lot of man power to input the coupon information into computer. Therefore airline companies have showed great interest to invest in automatic coupon recognition system.In order to satisfy the requirement of social application, we developed an automatic airline coupon recognition system. In our previous research, our system recognizes ticket-number (included Coupon Number, Serial Number and Check Digit) and airline-code automatically. And the system has been used in Shanghai Airline in China for two years. It has processed nearly 10 million coupons and gets a recognition rate more than 97%. But with the increasing of application requirement, the coupon fare also need to be recognized automatically, so that it can connect to the accounting system to produce some useful information.First, comparing with the background of Serial Number, the background of coupon fare is more complex. All of them are gray and have table lines. And the grayscale of background varies greatly in different flight coupons. What's more, the flight coupon fare is multi-font and the font size is unfixed. The multi-font increases the discrepancy of the same class and decreases the similarity of different classes. This is the main reason that leads to error recognition. In this case, the previous OCR model, which is a signal neural network, can not separate them form each other effectively and have a high error rate. So we need a robust OCR mode to finish this job. The novel solution is based on multi neural network which included back propagation neural network and convolutional neural network. We also use two layer OCR model to recognize some special character. The multi neural network system performs very well in fare recognition. |
PDF Full Text Request