An algorithm for ATM recognition of spliced money based on image features

According to the problem of spliced money in ATM (Automatic Teller Machine), the paper puts forward a template matching algorithm for recognition of left and right numbers based on one-dimensional gray scale projection. First it uses the Canny boundary detection algorithm to separate the banknote boundary from the R component image, and then preprocesses the image to make the subsequent processing easier; secondly applies the predefined location to divide the area of left and right numbers; thirdly calculates the one-dimensional gray scale projection curves of left and right numbers, based on which every character or number is divided; finally carries out the operation of template matching to obtain correlation coefficients of every character or number, comparing which with threshold it judges whether the left and right numbers are matched. For the transparent tape attached to the spliced money, a method based on the difference in object reflectivity to recognize the tape is designed. First it highlights the characteristics of the tape by preprocessing and binary morphological analysis of the B component image. Finally, it judges whether there is transparent tape on the surface of the banknote according to the set threshold value. Experimental results show that both number matching and tape detection can be completed, and algorithm can correctly recognize spliced money. The correct recognition rate of the algorithm reaches 100%, and the false positive rate、false negative rate are both 0. The processing is fast and simple, which meets the requirements of ATM.