Reseach On Marching-on In-time Scheme And The Fast Algorithm Of Time Domain Integral Equation

Invoice recognition system is a hot issue in current document analysis and recognition field, which includes image preprocessing, character segmentation and recognition. This paper focuses on the research of the identification of invoice numbers. A set of designs on the invoice number recognition system based on scanning pen have been developed, including image preprocessing, character localization and recognition. Simulation of the complete invoice recognition system based on MATLAB has been done. Eventually, the algorithm is transplanted into the hardware platform on scanning pen, which has achieved fast, accurate consequence.First of all, the invoice images collected from scanning pen have a variety of noise, and the illuminance is not that balanced. Then, the first step is preprocessing, which includes image denoising, illuminance balance and binarization. This paper adopts background estimation method calculating image background, which provides a good foundation for the following binarization, in case of the circumstance that some characters get broken after the binarization, morphological processing is adopted as a post processing method for improving, which has a very good effect.The collected invoice images, digital string is usually located in some particular location. There may some interference around, such as Chinese characters, seal, colon, border shading interference. while the subsequent recognition just points at digit string, so isolating the target string of numbers is necessary, this paper completes the rotation correction based on projection, then separating the digital string from the whole image according to the position of upper and lower bound.After a complete digital string bar obtained, further step is to segment single character. This paper presents a method solving connected string separation in complex environment, which has a good segmentation effect.This article has conducted the research of digit identification algorithms based on the research of common characteristics of the 10 digits. We put forward a method based on character structure, mainly character strokes and contour profile features, which obtains high accuracy and strong robustness.Based on the laboratory development platform, combining the previous relevant solutions, simulation and test have been completed in MATLAB. C language is used for scanning pen embedded software. We develop an applicable financial recognition system, which has been applied to the relevant business system. In this paper, about 97 invoice images collected from the scanning pen have been tested in the experiment, the recognition rate is up to 94.91% with a high real-time performance.