A Text Line Recognition Method in Chinese Scenes Based on Residual Convolutional and Recurrent Neural Networks

摘要

#$%^&*AU2020101229A420200806.pdf#####Abstract of Description The invention discloses a text line recognition method in Chinese scenes based on residual convolutional and recurrent neural networks, which comprises the following steps: collect text training images in Chinese scenes, perform normalization processing on the training image size, perform data augmentation processing on the training images, design residual convolutional neural network, residual recurrent neural network and CTC model, train horizontal text lines and vertical text lines and select the results with higher degree of confidence as the recognition results; the invention solves the problem of text line recognition in Chinese scenes by combining convolutional neural network and recurrent neural network, avoids the character segmentation of text lines and the error recognition caused by error segmentation, and accelerates the training of model by adding residual connection in convolutional neural network and recurrent neural network, so as to obtain a practical text recognition model in Chinese scenes. What's more, it has strong robustness and can recognize Chinese text lines with complex backgrounds, complex lighting and multiple fonts.Drawings of Description Start Collect text training images in Chinese scenes Perform normalization processing on the training image size Perform data augmentation processing on the training images Design deep neural network respectively to recognize horizontal and vertical text lines Obtain the recognition model by training a large number of training data Input the image to be recognized into the horizontal and vertical recognition models respectively, and select the results with higher degree of confidence as the recognition results End Figure 1 1/2