Research On The Printing Number Identification Algorithm In Financial Instruments

China's chromatic invoice layout is particularly complex,diverse,and some of its characters are very small.And the invoice is printed by the needle printer,resulting in a lot of invoice layout is not clear,skewed and distorted.There are many non-standard stamp and signature on the invoices.So how to correctly determine the location of the characters on the invoice image,how to correctly break up the characters with different sizes,how to determine the higher dimensional valid features of small characters and how to design the efficient authentication classifier are problems that have not been solved so far.This paper intrduces an effective invoice number recognition algorithm,which is based on the theory of image processing and pattern recognition,combined with the improved layout analysis and recognition technology.In the preprocessing phase,median filtering technology combined with a variety of filtering techniques to filter the image for removing salt and pepper noise.An improved method for tilt correction of white run length images is used for the oblique invoice images in this paper.Then the two value of the image is solved by iterative threshold method.According to the characteristics of the invoice and the analysis of the gray histogram,we can design the method of locating the invoice number.Finally,a single number is divided by the projection method,the template method is used to normalize the characters.In the feature extraction stage,the 40 characters are extracted from the printed number,which ensures that the small size number can be distinguished from other numbers.In the stage of number recognition,this paper proposes the improved SLAM classifier,which optimizes the original SLAM proposed by Mr.Wang and obtains a better classified result.In addition,in terms of the training sample library and the testing sample library with invoice number,the training sample library with 400 invoice numbers and the testing sample library with 300 invoice numbers are constructed by adding different noise,rotating angle and changing ratio for the original library that contain the training samples of 40 invoice numbers and the testing samples of 30 invoice numbers.The new SLAM classifier based on the improved invoice number library has pretty good recognition rate and antinoise performance.The experimental results show that in terms of rate of recogniting numbers,antinoise performance and recogniting speed,compared with the traditional BP network,the new classifier is better than BP network.