Year: 2024
Type: conference-paper
Abstract: This paper presents a technique for removing unnecessary QR code patterns from captured images of subjects using a U-Net type autoencoder. This study is a part of our series focusing on optical watermarking embedded invisibly in the light illuminating the subject. The purpose of optical watermarking is to include useful or valuable information about the subject in the captured images taken with a camera and provide it to the user as needed. We utilized QR codes as the watermarking in this study. A negative and a positive pattern are projected onto the subject alternately, making them invisible to the human eye. Although it is invisible to the human eye on the subject, the image taken by the camera contains positive or negative patterns, allowing users to extract information about the subject from the QR code in the captured image. However, if users want to save an image taken of a subject, the QR code is not necessary, and it is desirable to delete it after acquiring the necessary information, as it degrades the quality of the subject image. In this research, an image with a superimposed QR code on a subject image was created through simulation by multiplying the subject image and the QR code image. The autoencoder was then trained using the original subject image as the truth image. Using the trained model, we attempted to remove the QR code by inputting a subject image with a superimposed QR code that was not used during the training of the autoencoder. The evaluations of the difference from the truth image, measured using PSNR (Peak Signal to Noise Ratio) and SSIM (Structural Similarity), were 34 and 93, respectively, indicating that the QR code can be visually removed from the subject image.
Authors: Kazutake Uehira, Hiroshi Unno
Institution: Kanagawa Institute of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4394772016