Research On Aesthetic QR Code Algorithm Based On Gaussian Model And Image Style Transfer

QR code has become a part of daily life because of quick decoding rate,high information density,anti-rotation and other advantages.The conventional QR code has bad vision effect because it is constructed with random black-white modules.Many aesthetic algorithms of QR code have been proposed by global researchers.However,these algorithms fail to offer a good balance between the decoding rate and visional effect.An aesthetic algorithm of QR code based on decoder is proposed in this thesis,it improves the decoding rate and guarantees aesthetic vision effect.The sketch of image is also proposed,and two approaches of image style transfer based on feed-forward neural network is used to solve the problem of aesthetic algorithms of QR code with facial image.Main contributions are listed as follows:(1)An aesthetic algorithm of QR code based on the sampling of decoder is proposed due to the balance to the decoding rate and visional effect.The sampling of decoder is analyzed,a block binarization algorithm based on Gaussian distribution is proposed to the binarization of the background image provided by users.The results is used as the target image of modifying the original QR code in later stages.RS(Reed-Solomon)encoding is used to construct Positive Basis Vector Matrix(PBVM)and Reverse Basis Vector Matrix(RBVM).It is used to modify the color modules of the original QR code and the error correction ability of QR code is not impacted.The modified QR code looks like the background image.A fusion approach of grayscale QR code based on Gaussian distribution is proposed,the grayscale QR code is obtained after the fusion of QR code and the grayscale original image.The final color aesthetic QR code is obtained after the color rendering of grayscale QR code pixels and the background image.Experiment showed both higher decoding rate and good visional effect by the proposed aesthetic algorithm.(2)When facial image is used as the background image,the visional effect is not ideal and the decoding rate is low.The sketch of facial image and the image style transfer are proposed.The sketch of facial image is used to determine the borders,shades,to fusion with the texture image,to generate the sketch image.The DeepLab model is used in the facial image split.The fusion range of the image is under control,the sketch image after the fusion is more realistic.The color rendering is necessary for the visional effect because the sketch image is just grayscale.The current image style network model is improved,the facial image can transfer toward any style according to model,and the trueness of the facial image is preserved.The above two approaches are used in the algorithm's preprocessing of the original image.The facial image in the generated aesthetic QR code after such preprocessing has high decoding rate and good visional effect.