基于CNN和LSTM的自然场景文本检测应用

Application of the Natural Scene Text Detection Based on CNN and LSTM

针对自然场景的文本检测,构建了一种基于卷积神经网络(CNN)和长短时记忆网络(LSTM)的自然场景文本识别框架,运用CNN网络对图像中的静态特征进行提取,LSTM提取上下文特征信息。在解码上,提出了一种混合的CTC-Attention机制对输出层的编码进行解码。

Aiming at the text detection in natural scenes,a text recognition framework based on CNN and LSTM network is constructed.The static features in images are extracted by CNN network,and the context features are extracted by LSTM.In decoding,a hybrid CTC-Attention mechanism is proposed to decode the encoding at the output layer.