| In this society where intelligent identification technology and computer graphic information processing are developing rapidly,the halftone information hiding and information anti-counterfeiting technology in the field of printed information anti-counterfeiting has attracted more and more attention.At the same time,these two technologies have also developed into one of the mainstream technologies in the field of printed information anti-counterfeiting.In addition,due to the research and development of computer hardware in the past few years,solving problems in image processing through using deep learning has also become a major research direction.This paper is about the halftone image with pseudo-random noise information implanted,which is a QR Code with information hiding function and anti-copy performance.In the following paper,it will be referred to as a Secure QR Code.Since the size of the embedded and hidden information(noise image)in this type of QR Code is usually about 20-30 microns,it is usually difficult to achieve the required definition(signal-to-noise ratio)requirements for direct reading with a mobile phone.Furthermore,it also has problems,like high reading error rate,low reliability,and user experience is difficult to achieve satisfactory results.Therefore,in order to solve these problems,this paper will use neural networks and deep learning methods to restore the low-quality images scanned by mobile phones to generate high-quality images,and to read the hidden information embedded in the images.At the same time,it also creates a BP neural network to learn the non-linear mapping relationship between the blurred Secure QR Code image and the original QR Code images.After the model converges,restore the image by using the model.Then,this paper will perform high-quality restoration of low-quality images collected by mobile phones according to microscopic information’s characteristics which implanted in the Secure QR Code,and build a generative adversarial network to restore the microscopic information in the Secure QR Code image.This method further improves the scanned image quality of the mobile phone,and effectively solves the problem of poor scanned image quality.Finally,according to the black and white blocks of this type of QR Code and the characteristics of carrying the microscopic scale,this paper will divide this QR Code image into blocks,and classifies and labels the image blocks based on the embedded noise information.After that,setting up a convolutional neural network model to learn the microscopic information’s characteristics,which implanted in the Secure QR Code image,effectively reads the microscopic noise information.As the reading result,it can meet the error tolerance requirements of pseudo-random noise image information coding.Eventually,through experimental verification,the problem of Secure QR Code image restoration and carrying information reading has been completed,and making this type of QR Code technology more widely used. |
