Automatic Recognition Of QR Codes Under Complex Imaging Conditions

With the advent of the information age,two-dimensional code(QR code: quick response code)is widely used in logistics management,electronic payment,information integration and other fields because of its low price,high security and large information capacity.QR code recognition technology has also become a focus in the field of automatic recognition.To solve the problems of noise,uneven illumination,tilt and distortion in the process of QR code image acquisition,and to ensure the accuracy of QR code recognition,is the focus of this research.The detecting and alignment technology of two-dimensional codes are studied through theoretical analysis and experiments,a QR code recognition software with good recognition rate is built.The main work and innovation are as follows:Firstly,aiming at the problem of multi-code detection in single image,QR code detectors are trained based on artificial features and deep learning features respectively,and their performances in QR code detecting are analyzed and compared.For the artificial feature,the ability of haar features to describe the module of QR code is analyzed.By training cascaded QR code detectors,candidate regions containing QR code are extracted.For the learning feature,the mobileNet based on SSD multi-scale detection framework is used to extract the depth features,and the network is fine-tuned for model training and QR code detection.It is verified by experiments that the detection rate of the QR code detector obtained by the learning feature is higher than that of the single feature designed by manual,especially in the images with blurred,shadow and perspective transformation.Secondly,for different types of distortion QR code correction,image preprocessing is used to reduce the impact of noise in the first,an adaptive binarization method is proposed to obtain a clear QR code binary image,and based on this image,an algorithm by detecting the corner of QR code through Finder Pattern is proposed.The algorithm optimizes the traditional method that based on Finder Pattern,combines the contour features,and determines the distortion type by linear features of the key points.For perspective distortion,the four vertices of symbols are corrected by anti-perspective transformation;for nonlinear distortion,the mapping relationship between standard symbols and nonlinear distorted symbols is calculated by least square method based on the key points,and QR codes with nonlinear distortion are corrected by image interpolation algorithm.The experimental results show that the proposed algorithm is effective in correcting QR codes with perspective distortion and regular nonlinear distortion.In addition,combined with the open source zbar decoding algorithm,the QR code recognition software based on Visual C++ is designed and implemented,and the algorithm proposed in this paper is applied.The software mainly includes display interface,QR code detection and decoding module.By testing the performance of the software,it is proved that the QR code recognition software can be applied to complex environments.