Research Of Recognition Method On Invoice Printing Number

Invoice management work has always been done in a traditionally manual way for many years, which exists a series of problems such as arduous task or low efficiency because of the large invoice processing base and limited time. To solve this problem, this paper aims to use the computer to automatically process invoice, which could not only save labor and diminish resources consumption and capital investment but also improve working efficiency. The key work of invoice processing is to record the invoice number, therefore the emphasis of this study is to identify invoice number and invoice code through the computer, which means the recognition of printed numbers in ordinary commercial machine-printed invoices. Digital recognition is a kind of optical character recognition; we found that the simple theory research has matured through the comprehensive analysis of current research results in this field at home and abroad. However, the combination of theory and practical environment to specific digital identification technology still has great research value and space.Invoice number recognition need to take pictures of the invoice, photographic condition led to the image effects is not controllable, how to identification the digital information in the invoice image existence of interference factors is the difficulty of this paper. In this regard, the main work of this paper is as follows, first of all, the image of the collection is used for tilt correction and number location, the method which is based on Hough transform is adopted to detect the tilt angle to correct the invoice images to horizontal state, and the location of the digital area is realized by the projection method. Afterwards, we study the invoice image preprocessing, which includes image noise removal, image gray, binaryzation, character segmentation and normalization. Encoding implementation is undertaken for each preprocessing algorithm, using a nonlinear median filter for image denoising, the method of image binaryzation uses the adaptive threshold segmentation algorithm, randomly selecting a value as a start threshold value, and the final threshold value is determined by the calculation of continual iteration. Projection method is applied to image segmentation cutting the numeric string into single figures; Interpolation method is used to resize the single number to the same size. Finally, an analysis of several printed digital recognition algorithms is done. An improved algorithm based on the feature of digital structure is proposed due to its structure stability and morphology unicity.Comparing to the structure identification algorithm based on crossing number, the number recognition algorithm based on structure feature and the improved left and right contour feature recognition algorithm, the improved recognition algorithm based on the digital structure feature adapted in this essay has a relatively high recognition rate that is 98.5%, simultaneously having a low running time and a good noise robustness. The result suggests that the improved recognition algorithm based on the feature of digital structure can improve the invoice image recognition rate and accuracy. The implementation of this algorithm can effectively improve the working efficiency of the invoice processing and storage.