The Research And Implementation Of Banknote Multi-spectral Images Acquisition And Preprocessing Technology

Banknote authentication has been an important issue in the field of anti-counterfeiting, which is related to the financial stability of the country and has great theoretical significance and broad prospects for use. With the continuous development of digital image processing and pattern recognition, a brand new banknote authentication technology based on multi-spectral image has emerged and gained widespread attention. Multi-spectral image authentication techniques of banknote include the following four parts: multi-spectral image acquisition, image post-processing, serial number recognition and identification of authenticity. Multi-spectral image acquisition and processing are the premises of banknote classification and recognition, and the quality of banknote images directly affects the accuracy of counterfeiting and the performance of financial instruments. However, many factors may have effects on the quality of acquired images, such as instability in running speed of the machine, circuit system thermal noise, different degrees of tilt while running, the old and new of the paper currency, and so on. Therefore, the study on the multi-spectral image acquisition and post-processing technology of banknotes has great theoretical and practical value, as well as access to the standard and high quality banknote images.Aiming at the required technical indicators of financial instruments, this paper designed a high-speed multi-spectral image acquisition scheme based on CIS(Contact Image Censor),which is controlled by FPGA(Field Programmable Gate Array). FPGA can precisely control CIS to acquire images of the banknote in both sides simultaneously under multi-spectral, including white light, infrared light, ultra violet and other spectrum. And Ping-pong cache principle is used to achieve the high-speed transmission of multi-spectral images towards DSP(Digital Signal Processor). By analyzing the original image of banknotes and physical characteristics of CIS, we searched the noise source of images and came up with an image denoising algorithm based on the point by point compensation of CIS. And the algorithm has been implemented in FPGA hardware. For the random noise that may appear in images and the deformity of the banknote, this paper proposed a banknote positioning method based on edge extraction and parallel line detection. Considering the efficiency and image recognition feature regional distribution, this paper proposed a method of banknote image standardization through flexible use of various interpolation algorithm. Finally, in order to improve the accuracy and versatility of the algorithm, this paper put forward a bill direction recognition method based on image texture analysis. This method can not only identify the orientation of RMB accurately, it can also be applied to other currencies.The proposed multi-spectral image acquisition and post-processing method was examined and tested on certain platform with RMB. Compared with traditional methods,it has faster processing speed and higher image quality. What's more, it has a good adaptability for different types of banknote images, and a higher recognition rate of banknote orientation. The schemes and algorithms described in this paper have been achieved commercial production, and applied in a number of large banks.