Rsearch On Image Quality Improvement Method For OCR And Its Application

With the development of science and technology of the Internet,people hope to extract useful information in the paper by scanning and taking pictures,etc.Optical Character Recognition(OCR)technology is the key technology to solve this problem.The universal application of mobile devices such as mobile phones and cameras has brought people the convenience of taking pictures and obtaining information,but also brought many uncontrollable factors,such as partial exposure,blur,distortion,etc.,which have reduced the image quality.In turn,the OCR recognition rate is reduced,the image blurring and image distortion is the most common cause low recognition rate of two qualitative factors.In this paper,we study the text image quality improvement method under the deep learning framework on the above problems,and restore the degraded image to improve the image quality,and then improve the text recognition rate.The main contents are as follows;(1)In the problem of image deblurring,for the problem of poor applicability of different degrees of blurred images to convolutional networks,two image deblurring algorithms based on multiscale features are proposed:an image deblurring algorithm based on multiscale feature generation against the network.The generation model of the algorithm uses a multi-scale structure to solve the feature fusion problem and improve the problem of blurred image restoration at different levels;The adversarial model of the algorithm uses a penalty function with a penalty function to limit training to a controllable range and accelerate model convergence.Based on the multi-scale codec network image defuzzing algorithm,this algorithm proposes a multi-scale residual module under the multi-level codec structure,so as to improve the network's ability to defuzzing different degrees of blurred images under different sensing fields.In this paper,the two proposed deblurring algorithms are compared and tested in GoPro dataset and text dataset,and verify the effectiveness of the two algorithms.(2)An algorithm based on edge model and Hough line detection is proposed to solve the problem of edge extraction with much interference in complex environment.In the aspect of edge detection,multi-level and multi-scale edge detection model is used,and an image edge dataset with perspective distortion is constructed to reduce background edge interference and realize targeted edge extraction.Secondly,Hough Probabilistic Line detection is used to detect edge images,and a series of screening mechanisms are proposed to obtain edge lines and corner points in text areas,so as to achieve the restoration of the distorted text images.(3)Based on the above algorithm,an image quality improvement system for OCR is built.The system mainly includes the functions of local overexposure detection,text image deblurring,distorted image correction and so on.For the local overexposure problem,the brightness calculation method based on sliding window is used to detect.The comparison of the recognition rate of text images before and after processing shows the effectiveness of the text image quality improvement system.In summary,this paper proposes an image deblurring method based on multi-scale features for blurring images of different degrees,which effectively improves the clarity of the image,and an image correction method based on edge model and Hough line detection for image distortion,which effectively improves the distortion image correction.Accuracy,on this basis,build a text image quality improvement system,through the text recognition rate comparison verification,the system can achieve a significant increase in recognition accuracy.