High Capacity Adaptive Texture Synthesis Data Hiding Algorithm

• Main Authors: Chia-Shen Tai, 戴佳燊
• Other Authors: 王宗銘
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/nyw6ms

碩士 === 國立中興大學 === 資訊科學與工程學系 === 106 === Steganography realizes a method to communicate secretly by concealing secret message within digital media without drawing attention from eavesdroppers. Although texts, images, audio, video, 3D models are acceptable media to deploy steganography, we focus on the image-based steganography in this thesis. In 2015, Wu and Wang proposed a steganographic algorithm based on reversible texture synthesis. Their algorithm is known as SRTS (Steganography using Reversible Texture Synthesis). In contrast to traditional image steganography, SRTS does not need to modify the cover image; instead, SRTS generates a larger, synthesized image from the source texture. Utilizing the fact that there is no cover image or stego image in texture synthesis-based steganography, SRTS is able to resist attacks, such as machine learning-based steganalytic approaches. However, SRTS exhibits three weaknesses. First, SRTS provides a low embedding overall capacity, in comparison with traditional image steganography. Second, the embedding rate, measured in bits per patch, is a fixed value for each synthesized patch, instead of referring to a dynamic value determined from the context of each synthesized position. Finally, SRTS employs a mirroring technique in order to refer pixels out of the source image. As a result, the size of the patch is vulnerable to eavesdroppers, causing security issues. In this thesis, we propose an algorithm to resolve these drawbacks. Our algorithm, High Capacity Adaptive Texture Synthesis (HCATS), conceal secret messages in the source image prior to texture synthesis using a reversible embedding algorithm. Specifically, we combine optimal linear color transfer (OLCT) and reversible steganography exploiting modification direction (REMD) scheme as our first stage of embedding. Secondly, we introduce an adaptive texture synthesis scheme which determines the threshold from mean squared errors between an overlapping region of a position to be synthesized and that of the candidate patches. Increasing the number of acceptable candidate patch ensures our scheme can offer more embedding capacity without sacrificing the quality of the synthesized image. Lastly, we analyze the vulnerability of the patch size and propose countermeasures to resist against attacks of speculation about the patch size. These recommended counter measurements reinforce our algorithm, significantly improving the overall security. Experiment result shows that compared to SRTS, our algorithm increases the embedding capacity by the magnitude of 601%. In summary, we present a novel steganographic algorithm incorporating optimal linear color transfer, reversible steganographic embedding by exploiting modification direction, and adaptive texture synthesis. The proposed algorithm provides feature of source texture reversibility, increases the security to withstand malicious attacks, and offers high embedding capacity outperforming the current state-of-the-art schemes.