Jun 17, 2018
Pin Wu, Yang Yang, Xiaoqiang Li
StegNet: Advancements in Image Steganography Through Deep Convolutional Networks Abstract: Traditional image steganography, the science of encoding a secret image into a host image, has primarily focused on safely embedding hidden information while disregarding payload capacity — the volume of hidden data the host image can contain. The authors of this study utilize deep convolutional neural network methods to address this limitation, achieving a significant increase in payload capacity while maintaining the integrity of the hidden image. I. Introduction Image steganography, a cryptographic field, is centered on the practice of subtly modifying a host image to transfer hidden information, typically without third-party awareness. Historically, the main focus has been on safely embedding the hidden data, with payload capacity often left as a secondary point of consideration. Indeed, standard image steganography algorithms put significant emphasis on hiding information in the host, or cover image, without paying much mind to the volume of hidden information. Increasing payload yield is paramount to the advancement of steganography methods since more hidden information often means a greater alteration to the visual manifestation of the cover image and so, a higher risk of detection. Major traditional steganography methods, such as those hiding large data files during transmission by embedding a RAR archive after a JPEG file, can theoretically store infinite amounts of extra data, yet are highly vulnerable to any form of editing that disrupts the cover image. II. High-order Transformation To maximize payload capacity while resisting alterations, pixel-level steganography is often used. Dominant within this area are methods like Least Significant Bits (LSB), Bit Plane Complexity Segmentation (BPCS), and their extensions. However, while these methods can provide significant payload capacities, they are typically vulnerable to statistical analysis, and hence are easily detected. Moreover, they often make the hidden image vaguely visible. III. The Promise of Deep Convolutional Networks Deep Convolutional Neural Networks (CNN) make multi-level high-order transformations possible for image steganography. This method allows us to consider how to hide information as well as where to hide it, allowing for a remarkable payload capacity of 98.2% or approximately 23.57 bits per pixel (bpp) with only around 0.76% of the cover image altered. Also, the embedded image generated by this method proves robust against statistical analysis. IV. Autoencoder Neural Networks & Future Prospects The architecture of this study is inspired by classical auto-encoder neural networks, trained to produce an output image similar to the input image. A key contribution of this work is the implementation of CNN for image steganography, embedding image data without traditional steganographic methods. This area promises rich future explorations, particularly in denoising, dimension reduction, and image generation applications.
Sign up to AI First Newsletter
Morita Equivalence of C^*-Crossed Produc...
ByNandor Sieben
Oct 3, 2010
Characterizing Video Responses in Social...
ByFabricio Benevenuto,...
Apr 30, 2008
Toward Ethical Robotic Behavior in Human...
ByShengkang Chen, Vidu...
Jun 21, 2022
StegNet: Mega Image Steganography Capaci...
ByPin Wu, Yang Yang, X...