A Non-Bottleneck Residual Approach for QR Code

With the rapid development of the internet and the rising popularity of smartphones, especially mobile phone cameras, QR code has gained popularity outside their original use, from inventory tracking in factories & logistics to advertisements, electronic tickets, and even mobile payments in the commercial field, making them readily available to everywhere. With the advancement of deep learning, neural networks, and computer vision, many important statistical features can be automatically extracted and learned to improve identification accuracy. This research studied many CNN models for QR code-based printed source identification through a series of experiments that involved color QR codes to observe whether the bottleneck residual method is necessary for printed source identification. Our simplified residual model could compete with all of the tested models in color printer source identification even without bottleneck residual blocks implemented.