Research On Text Detection And Recognition In Natural Scene

Text can be seen everywhere in the natural environment.People can transmit and exchange information through words.Because the scene text contains important semantic information,it is very important to detect and recognize the scene text information accurately.The scene text detection and recognition technologies have important applications in intelligent transportation system,unmanned driving and other fields,especially in intelligent transportation system.With the rapid development of China’s economy,people’s living standards have been improved,and the number of car ownership has also risen sharply,which makes the traffic congestion seriously.As a part of intelligent transportation system,the text detection and recognition technologies of scene signs are studied in this thesis,which is based on the text recognition of traffic signs.The main work is as follows:(1)Analyzed the traditional scene text recognition technology OCR,pointed out the shortcomings of OCR recognition system in the scene text recognition.The related theories of deep learning are described,and the deep learning framework used in this thesis is studied.The scene text detection data set is analyzed and the traffic sign text detection data set is made.(2)The detection and recognition technology of text in traffic signs based on color and shape is analyzed and realized.The color features and shape features are fused,and the K-means color clustering algorithm is used to realize image segmentation.Then the text is detected by connected region analysis method.Finally,OCR recognition system is used to recognize the text.(3)A scene text detection method based on Mask R-CNN is proposed.This paper analyzed the deep learning Faster R-CNN algorithm,then studied the Mask R-CNN algorithm,proposed a scene text detection model based on the Mask R-CNN method,trained the data set using the Mask R-CNN text detection network model,obtained the scene text detection model,and trained and validated it on the data set.The experimental results show that the network model of this thesis can accurately detect the scene text.(4)For the scene text recognition technology,this thesis uses the idea of coderdecoder.CNN and BiLSTM are used to encode the scene text,and then Attention mechanism and CTC are combined to decode the scene text to complete the recognition.Finally,experiments are carried out to verify the effectiveness of the proposed method.