Anti-counterfeit Identification Technology Research in Banknotes Based on Multi-spectral Images
|School||Huazhong University of Science and Technology|
|Course||Communication and Information System|
|Keywords||Multispectral banknote image Banknote identify Feature extraction Kernal Fisherdiscriminant analysis|
Banknote multispectral image security identification technology has been the hot spotin the field of paper currency anti-counterfeiting, relating to the national financial security.The research has great theoretical research value and widespread application background.Banknote image contains the image information of visible light, infrared light and UVlight. Through analysis and understanding of the notes multispectral image, we canclassify these banknotes high reliably. Also we can stably detect the fake money that thetraditional anti-counterfeit technology is unable to distinguish, as well as the money thatare altered and in small circulation. Thus guarantee the security and reliability of thecurrency notes.Multispectral image security identification technology of paper money includes fourprocess, namely the image acquisition, preprocessing, denomination and directionrecognition, and authenticity identification. At present, researchers have proposed manyideas regarding to these processes and made good progress. But the main problem lies inthe stage of authenticity identification, where extracted features and the design ofclassifier can’t apply to different illumination, types, old and new, and noise pollution ofbanknote images. Its ability to detect the fake money is weak, leading to themisinformation of real currency.In this paper the authenticity identification stage, through the use of many regionalcontrast type strategy puts forward the multispectral image feature selection algorithmarea; By the use of the same spectrum of different block image effective gray informationof the relative relationship between, and puts forward the gray feature extractionalgorithms; By the use of different spectral the same block image matching relationbetween eight to structure, and puts forward the structure of grain feature extractionalgorithms; In order to improve the classification ability, this paper also introduced thesupervised training method, is proposed based on FDA classifier design algorithm. Thisalgorithm can solve the note image illumination differences, degree and type change, noisedisturbance, adapt to and to improve based on spectral image security appraisal systemstability. Based on the experiments on the euro and US dollar, this paper has proposed a newmethod. Compared with other methods, it can enhance to the compatibility of differentillumination, types, old and new, and noise pollution, raise the identification rate, and cutthe probability of misrecognition.