End-to-end tax bill text detection and recognition method

Abstract

The invention particularly relates to an end-to-end tax bill text detection and recognition method. The end-to-end tax bill text detection and recognition method comprises the following steps: firstly, roughly detecting a text region by adopting a target detection algorithm, proposing a boundary attention mechanism according to the characteristic of large spacing between certain characters of a tax bill text, and refining the text region; and then detecting a bent text in the tax stamp by utilizing a semantic segmentation technology, and performing text recognition at the same time. Accordingto the end-to-end tax bill text detection and identification method, horizontal and multi-directional texts can be detected and bent texts can also be detected by utilizing a detection and segmentation technology; meanwhile, the attention mechanism is used for detecting the text instances with large word spacing in an iterative mode, end-to-end detection and recognition are achieved, the accuracyof detection and recognition can be improved, and the efficiency of the whole process can also be improved.

Description

technical field [0001] The invention relates to the technical field of target detection and semantic classification in deep learning, and in particular to an end-to-end tax bill text detection and recognition method. Background technique [0002] In recent years, due to the huge application prospect of text detection and recognition technology (OCR technology), this technology has attracted more and more researchers' attention. Text detection and recognition technology can automatically and quickly obtain text information in different scenarios, which facilitates subsequent operations, such as handwritten document entry, scene understanding, automatic driving, geographic positioning, and many other aspects. [0003] The current work basically regards detection and recognition as two independent tasks-text detection task and text recognition task. The text image is first input to the detection model to locate the corresponding text area, and the text area is cut out and inpu...

AI Technical Summary

Problems solved by technology

Although some end-to-end designs have emerged recently, most of these models use a stacking strategy to accumulate detection and recognition modules in one model, resulting in an overall structure that is too complex and difficult to train

In addition, the current popular frameworks are basically based on English text, and there are very few models specifically for Chinese

Images