Method and device for generating handwritten character samples, electronic equipment, and storage medium

Abstract

The present application provides a method and device for generating handwritten character samples, an electronic device, and a storage medium. The method includes: acquiring a background picture without character texture and handwritten character pictures of different character types; Preprocess the handwritten character picture to obtain the first character mask map corresponding to the character type; determine the character area of ​​the handwritten character picture of the character type according to the first character mask map, and perform data enhancement on the character area to obtain the character type The corresponding second character mask image; randomly select the target character mask image from the first character mask image and the second character mask image corresponding to any character type, and fuse the target character mask image with the background image, Obtain handwritten character samples. In this way, a variety of handwritten character samples can be generated, and the diversified handwritten character samples can improve the generalization ability of the trained character recognition model and improve the accuracy of the model.

Description

technical field [0001] The present application relates to the technical field of natural language processing, and in particular to a method and device for constructing handwritten character samples, electronic equipment, and a computer-readable storage medium. Background technique [0002] In the commodity identification task in the retail scene, it is often necessary to identify the price tag and the corresponding price of the commodity at the same time. When the price tag to be identified is a supermarket standard machine-printed price tag or an electronic price tag, usually the price tag detection - OCR recognition technology can be used to effectively solve the problem, but when the price tag to be identified is a small supermarket or community retail store For handwritten price tags, the current handwritten character detection and recognition methods usually cannot achieve high recognition accuracy. The main reasons include: high diversity of handwritten characters (fon...

AI Technical Summary

Problems solved by technology

When the price tag to be identified is a supermarket standard machine-printed price tag or an electronic price tag, usually the price tag detection - OCR recognition technology can be used to effectively solve the problem, but when the price tag to be identified is a small supermarket or community retail store Handwritten price tags, the current handwritten character detection and recognition methods usually cannot achieve high recognition accuracy, the main reasons include: high diversity of handwritten characters (fonts, angles), high diversity of handwritten price tag backgrounds (different background color textures), The amount of handwritten character sample data is small, and the imbalance of different types of data is strong

[0003] Therefore, constructing a price tag recognition model training set only through online actual shooting data usually cannot achieve good generalization performance in actual use, and the recognition effect of some price tag scenes is poor, and it cannot be achieved by repeatedly supplementing actual data. Stable and rapid recognition accuracy improvement

Images