Scene text recognition method based on HRNet coding and double-branch decoding

Abstract

The invention discloses a scene text recognition method based on HRNet coding and double-branch decoding. When a traditional deep learning method is used for scene text recognition, the recognition accuracy is reduced when the problems of text distortion, image blurring and low resolution are encountered. The method comprises the following steps of: performing random Gaussian blur on an original text image of a single scene to obtain a low-resolution image; and building a scene text recognition model based on HRNet coding and double-branch decoding, wherein the scene text recognition model based on HRNet coding and double-branch decoding comprises a correction network TPS, a coding module, a super-branch branch and a recognition branch. According to the method, HRNet coding and double-branch decoding are introduced, the recognition accuracy of the model for fuzzy and low-resolution images is improved, and the model parameter quantity and time consumption are reduced in the mode of abandoning super branches during testing.

Description

technical field [0001] The invention relates to the technical field of computer vision and image text recognition, in particular to a scene text recognition method based on HRNet coding and double-branch decoding. Background technique [0002] Scene text recognition aims to automatically identify text content in natural scene images. Different from regular document text, text in natural scene images has the characteristics of changeable shape, complex background, distorted text, and blurred images. Early scene text recognition models were usually based on time series feature classification, using deep convolutional network VGG to extract image feature sequences, using recurrent neural network RNN ​​to learn bidirectional dependencies of feature sequences, and predicting the probability of text character sequences, and finally transcribing through continuous time series classification layer, which transcribes the sequence of predicted character probabilities into text accordi...

AI Technical Summary

Problems solved by technology

However, scene text images are usually disturbed by factors of different sources and varying degrees, such as complex backgrounds, text distortions, etc., which often cause information loss in the visual features extracted by the encoding network, which in turn causes the decoder to suffer from noisy decoding timestamps. Recognition of the target sequence is not accurate enough

Although the existing methods have a good effect on scene text recognition, they still cannot meet the accuracy requirements of scene text recognition.

Images