Handwritten document text detection and recognition method

Abstract

The invention particularly relates to a handwritten document text detection and recognition method. The handwritten document text detection and recognition method comprises two parts of text line positioning and text line detection. The text line positioning network uses a deformed VGG11 to perform network training on a picture, so that a possible starting position of a text line is found on the picture; and the text line detection network is incrementally propagated along the text line to obtain a text line starting position and a rotation angle, re-sampling is performed to obtain a viewing window, \the viewing window is input into the CNN network for regression to obtain a rotation angle of the next position until the rotation angle reaches the edge of the picture, and finally a normalized text line picture is generated and input into a text line identification network to recognize the text line picture and output a recognition result. According to the handwritten document text detection and recognition method, interference factors in a natural scene can be overcome, the text can be accurately detected and recognized, recursive advancing can be correctly carried out in the extension direction of the text line, and finally the bent text line is detected.

Description

technical field [0001] The invention relates to the field of deep learning technology, in particular to a method for detecting and recognizing handwritten document text. Background technique [0002] At the end of the twentieth century, the problem of position detection of text blocks in complex color images in natural scenes was first posed. The problem quickly became a hot topic in the fields of computer vision and document analysis due to the large economic and cultural benefits of solving it. In the decades since the above problem was raised, various text detection and recognition methods have been proposed. [0003] For text detection, there are mainly the following methods: [0004] 1. Based on the ability minimization method, most of the methods are based on conditional random fields and Markov random fields. The detection problem of text lines is regarded as an energy minimization problem to solve the interference between text lines; [0005] 2. The method based o...

AI Technical Summary

Problems solved by technology

However, due to the many interference factors in the handwritten document pictures taken in natural scenes, traditional text detection and recognition methods cannot be well applied to natural scenes.

Images