Auto ROI & mask R-CNN model for QR code beautification (ARM-QR)

The development of the Internet has enabled the QR code to become the most frequently applied two-dimensional barcode in daily life and in commercial advertisements, and its application continues to be more diversified to include warehouse management, electronic tickets, mobile payments, etc. The standard QR code consists of black and white modules, which display a monotonous visual effect. Since graph patterns are much easier to understand than text characters, showing the subject by patterns inside the QR code is the easiest way to understand implicit content. This research involves the development of a methodology called ARM-QR, in which the QR code is integrated with full-color images, and deep learning technology is used to beautify it. First, the region of interest (ROI) of the color image is automatically identified using Mask R-CNN. The QR code’s visual beautification is further adjusted by the content of the object. Discrete wavelet transform and contrast sensitivity functions are also used to strengthen the visual perception of the QR code and reduce the impact of a low print resolution on the graphic legibility. The ARM-QR code’s visual quality is intensively verified by visual quality indices, which include the Peak Signal-to-Noise Ratio (PSNR), Mean-Square Error (MSE), Structural Similarity Index Metric (SSIM), and Gradient Magnitude Similarity Deviation (GMSD) based on evaluating the experimental data. The results of the experiment confirm that the visual beautification of the QR code generated in this research is of higher quality than that in other QR code beautification studies.