The Research Of Banknote’s Image Recognition And Graffiti Detection Technology

With the development of economic globalization and trade integration, the recognition of multiple nations’ banknotes image has become a central issue in the field of banknote sorter. Using more efficient algorithms and multi-currency recognition will be the tendency in the future. This paper mainly studies the multiple nations’ banknotes image recognition and processing technologies, and discussing around the practical application of this technology.This paper studies multiple nations’ banknotes image recognition technology and comes up with the feature extraction method which combines global information with local information. It extracts three different kinds of image feature representing different scales, uses Bayes classifier to do the recognition and provides a pratical simplified algorithm. This method is not sensitive to the position offset of banknote and having achieved good result in whether a specific nation’s banknotes or mixed multiple nations’ banknotes recognition.Considering the high cost of getting some nation’s banknotes and the insufficient of training samples, this paper studies the technology of banknote degradation simulation. It analyzes the degradation situation of banknotes, proposes the simulation method of banknote degeneration which based on stochastic growth model and generating a large number of old banknotes using new banknotes which extends the training set. This method can simulate the degeneration of banknote in a better way, increase the number of training set samples, and effectively improve recognition rate.This paper also studied the detection technology of banknote’s graffito. It comes up with the method that using Gaussian second-order filter to do banknote image filtering, which can extract the feature of graffiti and enhance it. For ease of practical implementation, it uses the center line template of Haar-like to approximate Gaussian second-order filter. In order to extract graffito of different direction and thickness, it use the Haar filter template from two scales of four directions.