SSRH: screen-shooting robust hyperlink based on deep learning

Guangyong Gao

doi:10.1007/s00530-025-01689-6

2025-02-05

PRE

高

高光勇 (Guangyong Gao) *

Xiaoan Chen

Li Li

DOI：10.1007/s00530-025-01689-6

原文链接

原文求助

摘要

En 中文

With the development of informatization in recent years, the emergence of QR codes has opened up new paths in mobile payments. Some previous works have attempted to hide information in color images, called image codes, to solve the problem of QR codes not having visual semantics. However, the existing schemes have a low rate of correct information extraction in the face of screen shooting, especially under the interference of strong moir & eacute; effects. Therefore, the vision of this paper is to propose an image code method that can resist the moir & eacute; effect to improve the practicality of image codes in real-world scenarios. This work introduces a novel model called Screen-Shooting Robust Hyperlink (SSRH) designed to embed hyperlink information into natural color images, and the encoded images are invisible to the human eyes. At the same time, a screen-shooting distortion submodel is constructed between the encoder and decoder, which makes the SSRH model more robust against screen-shooting distortion in the physical environment. The SSRH separates screen-shooting noise simulation training from adversarial network training as a submodel, independent of the adversarial training, and utilizes Bayer CFA (Color Filter Array) to improve the generation of the moir & eacute; pattern. The experimental results provide compelling evidence that our proposed SSRH model surpasses state-of-the-art methods regarding its resilience against screen-shooting distortions, particularly those incorporating moir & eacute; distortion. The test code of the proposed method is available at https://github.com/Dawn-Hsiao/SSRH.

Keyword:

Data hiding

Robust hyperlink

Screen-shooting submodel

Adversarial training

Visual image codes