Identifying method of value-added tax invoice

Abstract

The invention provides an identifying method of a value-added tax invoice. The method is characterized by including the following steps: pre-processing an input invoice image; conducting option zone positioning on the invoice image; carrying out character division; dividing a character zone into nine sections; extracting the HOG feature of each image block, and obtaining nine one-dimensional HOG feature sequences; comparing the feature sequences with a template, and obtaining difference coefficients; on the basis of the difference coefficients, giving different weights to the nine sections and generating a weight matrix; conducting weighted summation on the weights and the difference coefficients, obtaining candidate character difference scores, calculating each candidate character difference score, and taking the character with the minimum value as a matching result; and repeatedly conducting the above steps on each divided character till all contents of the invoice are identified. The method effectively achieves Chinese character module identification in the value-added tax invoice, has an enhanced capability of distinguishing Chinese characters with similar forms, exhibits illumination invariance and rotation invariance, exhibits a higher anti-interference capability, and has a good capability of identifying fuzzy characters.

Description

technical field [0001] The invention relates to a method for identifying value-added tax invoices, and has a relatively high ability to distinguish characters with similar shapes. Background technique [0002] In the prior art, the identification of value-added tax invoices usually adopts the following methods: customize the form template, set the identification area, identify attributes, call special characters, identify the option area, perform identification post-processing according to the identification attributes, and finally output structured The recognition result; or on the basis of using Tsinghua TH-OCR technology, perform multiple preprocessing operations on the invoice, including deviation correction, color cast correction, color filtering, noise reduction, binarization, and enhancement of the contrast of the recognition unit. These functions can be flexibly configured and freely combined to output the best image quality for subsequent recognition. [0003] Curr...

AI Technical Summary

Problems solved by technology

[0004] 1. At present, only numbers are recognized in value-added tax invoice recognition, and there is a lack of recognition modules for Chinese characters in value-added tax invoices;

[0005] 2. The ability to distinguish similar characters is relatively weak, and misrecognition is prone to occur;

[0006] 3. The value-added tax invoice is made of carbon paper, and the characters and strokes are easy to stick when printing, which makes it difficult to identify

Images