Natural Scene Sext Detection And Recognition Method Based On Deep Learning

The research of natural scene text detection and recognition is an important research direction in the field of computer vision.Nowadays,the detection and recognition technology of regular scanned text has reached the advanced level,but for the text detection and recognition in natural scenes,due to the high complexity of font itself and background in natural scenes,the research on the detection and recognition of natural scenes still faces technical problems,which is a challenging subject.The main contents of this paper are as follows:(1)Through the analysis of the complexity and diversity of natural scene text itself,such as distortion,multi-color,font blur and so on,the background is also complex and diverse,such as background color diversity,low image resolution,background occlusion text and so on.In view of the high complexity of the natural scene itself,this paper proposes a natural scene text detection and recognition algorithm based on deep learning,which can detect and recognize the natural scene text in multiple directions and scales.(2)According to the characteristics of the natural scene text,such as uncertain direction and different scale,a natural scene text detection method based on multi-directional adaptive sensing region candidate network is proposed.By improving the anchor mechanism and introducing a multi-directional ROI align pool layer,the method can detect the natural scene text in multi-directional and multi-scale at the same time,with high detection accuracy and efficiency.(3)In view of the high complexity of text and background in natural scene,which makes it difficult to recognize,a text recognition method based on connection time classification and attention mechanism is proposed.In this method,the problem of text recognition in natural scene is transformed into a problem of sequence recognition,which avoids the disadvantage that the difficulty of character segmentation leads to the degradation of overall recognition performance.At the same time,the attention mechanism can reduce the network complexity and improve the recognition accuracy.(4)The structure of the text detection and recognition method for natural scenes is verified to the effect of the model in this paper.Firstly,the overall detection and recognition effect of the model is verified;secondly,the natural scene text detection module and recognition module are respectively modularized.Finally,through the verification on ICDAR13,ICDAR15,RCWT-500 and other data sets,it is proved that the proposed natural scene text detection and recognition method can effectively complete the task of natural scene text detection and recognition.In addition,a natural scene text detection and recognition system is designed.By inputting the scene text image,the visual detection and recognition of the image is realized.