Scene text recognition method based on generative adversarial style migration

Abstract

The invention belongs to the field of scene text recognition, and relates to a scene text recognition method based on generative adversarial style migration, which is implemented by a processor executing a program instruction and comprises the following steps: in a preheating training stage, implementing supervised learning training on a scene text recognition network by using a first scene text picture data set; in the synchronous training stage, setting a co-training network at the input end of the scene text recognition network, and using a second scene text picture data set for conducting synchronous training based on generative adversarial on the co-training network and an overall network formed by the co-training network and the scene text recognition network, wherein the co-training network comprises a generator for providing scene text pictures for the scene text recognition network; according to the accuracy of the scene text recognition network on the verification set after synchronous training, selecting model parameters of the scene text recognition network, and using the scene text recognition network to recognize text information of the scene text picture under the model parameters. According to the method, the scene identification problem under the condition of less real data can be effectively solved.

Description

technical field [0001] The invention belongs to the field of scene text recognition, and in particular relates to a method for text image recognition using a generative confrontational synthesis method. Background technique [0002] Scene text recognition is an important challenge in the field of computer vision, whose task is to automatically detect and recognize text in natural images. With the help of text detection and recognition technology, important semantic information in visual images can be decoded. Due to the huge application value of scene text recognition, it has attracted many people's research and exploration in the industry and academia in recent years. The training data used by existing scene text recognition methods are all synthetic data. Synthetic data has the advantages of large data volume, accurate labeling, and good scalability. However, due to certain differences with real data, it affects the performance of training using synthetic data. Content...

AI Technical Summary

Problems solved by technology

However, pure data augmentation still cannot directly solve the training problem of scene text recognition network

Images