TEXT RECOGNITION METHOD FOR NATURAL SCENE, STORAGE APPARATUS, AND COMPUTER DEVICE

摘要

A text recognition method for a natural scene, a storage apparatus, and a computer device. The text recognition method for the natural scene comprises: performing, by using a deep convolutional network, feature extraction on an image to be recognized to obtain multiple feature vectors (S11); encoding the multiple feature vectors by using an encoding two-dimensional recursive network combined with a residual module to obtain an encoded feature sequence (S12); decoding the encoded feature sequence by using a decoding two-dimensional recursive network to obtain a prediction label sequence (S13); and correcting the prediction label sequence by using a connection time classification model to form an output text sequence (S14). By means of the approach, the text recognition accuracy can be improved, and the performance of the text recognition method is improved.