Research On OCR Detection And Recognition Technology Based On Deep Learning

In recent years,with the rapid development of deep learning technology and neural network,image processing technology in the field of computer vision has also made great progress,which has promoted the progress of scientific research achievements of many domestic and foreign researchers.At present,text recognition technology is widely used in human-computer interaction,industrial automation,license plate recognition,banking,medical and other fields.At the same time,text is also an important channel of information exchange.Traditional text recognition methods have great problems in detection speed and accuracy,and cannot identify text information in complex scenes.Based on deep learning methods can solve this problem.Aiming at the problems of various fonts,inconsistent shapes and sizes,and large amount of small text,this paper proposes a text detection algorithm and CRNNs text recognition algorithm based on augment feature pyramid network(A-FPN)and IAM-Res Net network structure,and implements a set of deep learning-based text detection and recognition system.The main research achievements of this paper are as follows:(1)Aiming at the problems of text-like pixel false positives and small-scale text missed detection,A-FPN algorithm was proposed.The algorithm connected a feature augment algorithm(FAA)module in the high-level of the feature pyramid network(FPN),which made full use of the high-level semantic information.To solve the problem of rough text location information,an IAM-Res Net network structure was proposed.In this network,CBAM attention mechanism was introduced into Res Net50 backbone network to enhance the extraction of text feature information and improve the flow of text information between contexts.Compared with the existing algorithms,the algorithm proposed in this paper has improved the detection accuracy greatly and has excellent performance.(2)For character recognition feature extraction,this paper proposes CRNNs character recognition algorithm.The algorithm uses Bi-directional Long Short-Term Memory(Bi LSTM)to generate text sequence features from convolution features,and introduces the transcription layer into the attention mechanism for recognition model training during decoding.The experimental results show that the improved algorithm has some improvement in recognition accuracy.(3)Developed an OCR text detection and recognition system.For the development of PC system,the OCR text detection and recognition function is integrated together.The system provides users with two ways to select and recognize images,such as selecting local images and capturing local text images on the screen.After inputting the images,the images are transferred to the text detection and recognition model,and the detection and recognition results are displayed in the system interface.Users can save the detection and recognition results.Convenient follow-up check and modify the content of the text recognition.