Summary
International Workshop on Smart Info-Media Systems in Asia
2021
Session Number:RS1
Session:
Number:RS1-3
Method of Generating Pseudo-Captured Images to Evaluate the Performance of Data Embedding Techniques for Printed Images Using Mobile Devices
Masahiro Yasuda, Mitsuji Muneyasu, Soh Yoshida,
pp.28-33
Publication Date:2021/9/20
Online ISSN:2188-5079
PDF download
Summary:
Techniques to embedding data for printed images have been proposed. In these techniques, the embedded data is retrieved from the image captured using the camera of a mobile device. Although these methods are expected to be an alternative to QR codes, there are some problems, such as the data amount of embedding information and the robustness of detection. However, the evaluation of the methods requires printing and capturing actual embedded images, which is very burdensome. In this paper, we propose a method of reducing the workload in evaluating the performance of data embedding algorithms by simulating the degradation caused by printing and capturing. In this paper, we adopt a deep-learning-based approach to reproduce the degradation using Pix2Pix, an image transformation method based on conditional Generative Adversarial Network (GAN). In addition, we obtain the locations of the frames and the amount of distortion of the data embedded images from the captured actual images. These data from the captured actual images are utilized to cope with various captured conditions. Experimental results show that the proposed method can reproduce the detection accuracy of the embedded data under actual conditions and confirm the effectiveness of the proposed method.