Research Of RMB Banknotes Feature Extraction And Recognition Based On Image Processing

Along with the rapidly development of the social and the economic, more andmore industries and areas achieved of information technology and automatic. Fullyautomated service will be inevitable trend of development of the banking sector.Intelligent banknote detaching machine will be reflected best in the banking businessautomation.At present, the domestic banknotes sorter technology has some defects,such as the existence of the classification accuracy is poor, vulnerable to environmental constraints and poor ability to pick the residual. In this paper, based oncharacteristics of image recognition technology, I made a deep research and analysisto the banknote processing conducted.This paper mainly uses the current fourth and fifth edition of the RMB as ampleimage. In previous study, the collected banknotes image is pre-processed. Thenanalyzes the image preprocessing algorithms, including image enhancement, tiltcorrection and image segmentation of three parts. Methods using linear enhancementand spatial filtering enhancement of the image contrast enhancement. For Image tiltcorrection, using the multistage Hough transform: using coarse-to-fine search stepangle of inclination of the image detection, and the tilt angle of rotation correctionusing bilinear interpolation method.Image segmentation algorithm based on regionallocation, locate a rough area of the serial number.Identification of money, there are mainly two kinds of algorithms: analysis,comparing the one-dimensional and two-dimensional gray-scale projection algorithmtemplate matching algorithm. According to the characteristics of different algorithms,I used the matching algorithm based on one-dimensional gray-scale projection. Thisalgorithm is small,the main principle is: Firstly, transformed the image fromhorizontal projection and vertical projection; and then, changed two-dimensionalimage into a one-dimensional image; finally, the matching experiments. Experimentswere conducted to obtain the recognition rate of more than95%, the result isreasonable and reliable. Meanwhile, in the known version and the denomination ofthe bill, According to the establishment of different notes and training the neuralnetwork improved, We can achieve the recognition of paper currency image four oriented, the recognition rate is higher than95%.About currency serial number recognition, by repeatedly using the projectionmethod to achieved serial number character segmentation; by pixel by pixel featureextraction method extracted gray feature of letters and numbers, building a networkmodel. Using two different BP neural network to identify letters and numbers,finished the identification of the serial number. The identification rate of the numbersand letters above96%in this system.Finally, through the method of paper currency image histogram analysis andthreshold segmentation based on the old and new damage identification and notes.The results showed that the banknote image recognition algorithm of this paperto meet the relevant technical requirements of banknote sorting.