A method and device for recognizing characters and banknote serial numbers, and an ATM machine

Abstract

The invention provides a character image identification method for identifying a character image by adopting a vector characteristic automatic extraction network and a BP network, and meanwhile provides a character image identification based paper money crown word number identification method, a device for realizing the character identification method and an ATM (Automatic Teller machine) with a crown word number identification capability. The method and device for identifying the crown word number of the paper money and the character achieve the identification rate of above 99.6% for extracting and automatically identifying the crown word number of the paper money when being applied to a money depositing and withdrawing process of the ATM, and the unidentified part is mainly caused by the problems such as contamination of the paper money number, folding of the paper money number, inseparable number and background caused by the too old paper money, and the like.

Description

technical field [0001] The invention relates to the field of pattern recognition, in particular to character recognition, and relates to a double network character recognition method using a feature vector automatic extraction network and a combination of a recognition network and a device using the method. Background technique [0002] In the field of computer and telecommunications, a character is an information unit. For natural languages ​​that use alphabetic or syllabic scripts, it roughly corresponds to a phoneme, phoneme-like unit or symbol. Examples of characters are: letters, number systems or punctuation marks. At present, it is very common to use a string of characters (character strings) to represent identity features, and it also provides a broad application space for the automatic character recognition technology. For example, there is a string of unique strings on the currency, that is, the serial number, and there are checking account identification records...

AI Technical Summary

Problems solved by technology

However, among these methods, they all revolve around a core issue: how to design a better classifier or recognizer to improve the number recognition rate

[0006] Among these methods, it is found that although using the network as a classifier or recognizer to identify numbers is better, but at the same time, it is found that using a single network as a classifier or recognizer will easily cause the network prediction ability to overfit and be difficult to control. Grasping this nonlinear mapping makes the network structure too complex, resulting in the storage of a large amount of redundant information in the hidden layer, which greatly increases the load capacity of the network and affects the recognition accuracy.

Images