Scene text recognition-oriented super-resolution image generation method

Abstract

The invention discloses a scene text recognition-oriented super-resolution image generation method, which comprises the following steps of: firstly, collecting a plurality of image pairs, extracting binary mask images of the images and carrying out channel splicing on the binary mask images and an original image to obtain a group of image pairs, namely a group of training samples, wherein each image pair comprises a low-resolution image and a high-resolution image; constructing an image super-resolution network for scene text recognition, introducing a correction and alignment module into the network to preprocess a low-resolution image, adopting a training sample to train the image super-resolution network, and when super-resolution image generation needs to be performed on a scene text image, extracting a binary mask image from the low-resolution scene text image, splicing the binary mask image to obtain the scene text image, and inputting the scene text image into an image super-resolution network to generate a super-resolution image. According to the method, the image super-resolution network is improved, and the quality of the generated super-resolution text image is improved, so that the scene text recognition accuracy is further improved.

Description

technical field [0001] The invention belongs to the technical field of scene text recognition STR, and more specifically, relates to a super-resolution image generation method for scene text recognition. Background technique [0002] Scene Text Recognition (STR) is a popular direction in the field of computer vision. It has many applications in automatic navigation, image retrieval, human-computer interaction and other fields. With the popularization of photographing equipment such as mobile phones and cameras, it is more and more convenient for people to obtain images of natural scenes, and the demand for correct recognition of text information in images is also becoming stronger and stronger. In recent years, although the rapid development of deep learning has greatly promoted the development of scene text recognition technology, the recognition of low-resolution blurred scene text images has always been a difficulty and pain point in this field, which needs further resear...

AI Technical Summary

Problems solved by technology

In recent years, although the rapid development of deep learning has greatly promoted the development of scene text recognition technology, the recognition of low-resolution blurred scene text images has always been a difficulty and pain point in this field, and further research and improvement are needed.

Images