Chinese character OCR method and system based on neural network, medium and application

Abstract

The invention discloses a Chinese character OCR method and system based on a neural network, a medium and application, and relates to the technical field of image recognition. The method includes: performing text detection on the input image; performing text mask information extraction on the detected text area image, positioning the position coordinates of each character in the text area, and intercepting each character mask from the text mask according to the character position coordinates; performing feature extraction on the character mask, and mapping the feature to a multi-dimensional vector space to obtain a character vector; enabling the character vectors to be grouped into a text vector matrix according to a character order, and utilizing the text vector matrix for learning context dependence information to complete text recognition. According to the method, the problems of Chinese text detection of the input image, image text area interception and text area character recognition are solved. The Chinese character recognition accuracy is improved, the recognition accuracy of characters with similar shapes is improved, and the direction of the character is insensitive.

Description

technical field [0001] The invention belongs to the technical field of image recognition, and in particular relates to a neural network-based Chinese character OCR recognition method, system, medium and application. Among them, OCR is Optical Character Recognition, optical character recognition, that is, the text content on pictures and photos is directly converted into editable text. Background technique [0002] At present, in image recognition technology, there are convolutional neural network (CNN), text detection model EAST (Efficient and Accuracy Scene Text), U-Net neural network, long and short time neural network (LSTM), VGG (Visual Geometry Group ) neural network model. [0003] Scene text recognition aims to extract text content from images, which has great commercial value in practical applications. Scene text recognition usually includes scene text positioning and text recognition. For example, in Chinese text recognition, there are already a large number of m...

AI Technical Summary

Problems solved by technology

[0010] (1) In the prior art, the recognition accuracy rate of long character strings decreases; the position coordinates of each character cannot be obtained; the character direction is sensitive

[0011] (2) Chinese text detection cannot be accurately performed on the input image, and the text area of ​​​​the image cannot be intercepted, resulting in poor accuracy of text recognition in the text area

The average English word is composed of 8 letters. It is a big overhead to complete the positioning and recognition of each letter, and the algorithm is inefficient. This framework is a solution to the English language characteristics that simple letters form words and words form text. Recognize characters by identifying local features of characters and connecting contextual semantic features, the accuracy of each letter recognition is not so important

However, this method is not suitable for the recognition of Chinese characters

The characteristic of Chinese is that each character is an independent individual, and each character has the value of positioning and recognition. Compared with English, the segmentation between characters is relatively simple; Chinese has many similar characters, and this recognition method based on English It cannot cope with the distinction between shapes and near characters, so this recognition method has natural defects in Chinese recognition

Images