A Study of Optimal Adaptive Data Hiding Algorithms

• Main Authors: Siang-Yu Jiang, 江翔宇
• Other Authors: Chung-Ming Wang
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/36183796115643871326

碩士 === 國立中興大學 === 資訊科學與工程學系所 === 100 === This paper presents two adaptive image data hiding algorithms. The first algorithm we propose is an optimal, adaptive data hiding algorithm using the multiple base embedding technique to convey secret messages in grayscale/color images. We introduce a theoretical analysis which proves that adopting two levels of range table produces the minimal image distortion and achieves the optimal data embedding. Experimental results indicate that our method can generate a stego image that shows the optimal data embedding performance and provides an average of 0.38 dB higher PSNR values higher than our counterparts. Our algorithm is able to dynamically construct the range table, allowing us to satisfy user’s demand for the embedded capacity. Finally, a skill referred to as “circular shift” is presented to alleviate the abnormal fluctuation appearing in the histogram of stego image, thus increasing the robustness of the steganalysis attacks. This novel algorithm contains features of achieving the optimization, complying with user’s capacity demanding, and improving the robustness for the steganalysis attack. The second algorithm we propose is an optimized, adaptive data hiding algorithm for high dynamic range images. Similar to the first algorithm, our second algorithm dynamically constructs a two-level range table to satisfy the needs of the embedded capacity given by a user. A multiple base scheme is adopted to convey secret messages to achieve the optimization of adaptive data hiding. We propose a circular shift technique to alleviate the abnormal fluctuation appearing in the histogram of the stego image. In addition, we introduce a scheme referred to as an average base adjustment in order to ensure the legality of a stego pixel and provide robustness to resist the steganalysis attack. The visual difference predictor 2 (VDP-2) is adopted to measure the visual difference between images before and after the embedding. The output image concealing with the capacity of 12.9 bpp shows with a good image quality, where the tone mapped image has the PSNR value over 31.57 dB. Visualizing by the human eyes, there is an insignificantly low probability to perceive visual difference between cover and stego images. The proposed algorithm provides five remarkable benefits extending the feasibility of image-based adaptive data hiding for high dynamic range images. In conclusion, we propose two adaptive data hiding algorithms in this work. Experimental results confirm their superiority for the grayscale, color images, and high dynamic range images. We believe two novel algorithms provided contribute the applications of image adaptive data hiding.