Text recognition method and system

Abstract

The invention discloses a text recognition method and system. The method comprises the following steps: collecting a natural scene text image, inputting the text image into a text recognition model, and outputting a text recognition result corresponding to the text image, the text recognition model is obtained by training based on to-be-recognized text image samples and predetermined text labels, the text labels are in one-to-one correspondence with the to-be-recognized text image samples, and the text recognition model comprises an encoder and a decoder; the encoder comprises a conversion layer, a ResNet feature extraction layer and an optimization module layer comprising a first AOA module which are connected in sequence, the decoder comprises a BiLSTM sequence modeling layer and an LSTM prediction layer comprising a second AOA module which are connected in sequence, by introducing the AOA module, irrelevant or misguided attention results are filtered out, only useful information is reserved, and the text recognition accuracy is improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and more specifically relates to a text recognition method and system. Background technique [0002] Images in natural scenes contain rich text information. How to use technical means to recognize text in images, so as to provide convenience for people's lives, has become a current research hotspot. [0003] The current mainstream approach is to design an end-to-end neural network model with an encoder-decoder structure for natural scene text recognition. In the encoder-decoder architecture, the encoder uses a convolutional neural network to extract features from text images to obtain feature vectors, and then decodes them into corresponding text sequences through a network based on RNN (Recurrent Neural Network, cyclic neural network). Among them, the attention mechanism generates a weighted average of the extracted feature vectors at each time step to guide the decoding process of text...

AI Technical Summary

Problems solved by technology

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a text recognition method and system, the purpose of which is to filter out irrelevant or misleading attention results, thereby solving the technical problem of inaccurate text recognition

Images