Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (4): 235-242.DOI: 10.3778/j.issn.1002-8331.2109-0338

• Graphics and Image Processing • Previous Articles     Next Articles

Hiding Two Images with High Visual Quality

HU Xinjue, FU Zhangjie   

  1. 1.School of Computer Science, Nanjing University of Information Science and Technology, Nanjing 210044, China
    2.Engineering Research Center of Digital Forensics, Ministry of Education, Nanjing University of Information Science and Technology, Nanjing 210044, China
  • Online:2023-02-15 Published:2023-02-15



  1. 1.南京信息工程大学 计算机学院、网络空间安全学院,南京 210044
    2.南京信息工程大学 数字取证教育部工程研究中心,南京 210044

Abstract: To solve the problem that the current image steganography system cannot balance the payload capacity and imperceptibility well, this paper proposes a high visual quality method of hiding two image based on feature rich branch network. In the encoder network, the Res2Net-Inception module is designed to improve the feature extraction capability. In the decoder network, the W-Net network structure is designed to capture fine-grained details and coarse grained semantics from full scales. In order to associate with the human visual system better, a new mixed loss function which is more suitable for steganography is introduced to train the whole model. Experimental results show that the model effectively improves the image quality of both container images and reconstructed secret images while guaranteeing the hiding capacity compared to other image steganography models.

Key words: hiding two images, Res2Net-Inception module, deep learning

摘要: 针对当前以图藏图技术无法较好平衡隐藏容量和图像质量的问题,提出基于富特征分支网络的高图像质量的一图藏两图方法。在编码网络中,设计Res2Net-Inception模块提高特征提取能力。在解码网络中,设计W-Net网络结构,实现解码网络对图像细粒度细节和粗粒度语义的全尺度捕捉。为了更好地与人类视觉系统相关联,引入更适合隐写的混合损失函数训练整个模型。实验结果表明,相比其他以图藏图模型,该模型在保证隐藏容量的同时,有效提高了含密图像及重构秘密图像的质量。

关键词: 一图藏两图, Res2Net-Inception模块, 深度学习