The Study of Reversible Data Hiding Mechanisms

• Main Authors: Chang-Han Wu, 吳昌翰
• Other Authors: 張軒彬
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/3b588b

博士 === 國立中興大學 === 資訊科學與工程學系 === 106 === Data hiding is a scheme that embeds secret data in cover media. With the scheme, hackers cannot efficiently detect the secret data from the stego-media. Over the past decade, many techniques have been proposed for data hiding. However, most of them suffer from the problem of trade-off between payload and image distortion. There are four types of reversible data embedding methods, such as difference expansion, histogram shifting, prediction error expansion, and dual-image hiding scheme. The dual-image hiding scheme is the one that could generate slightly image distortion and avoids the underflow and overflow problems. Based on dual-image hiding scheme, Center Folding Strategy (CFS) takes the median to reduce each decimal digit for decreasing the modification level. Thus, embedding the transferred digits into the cover image does not lead to the acute distortion of the image. However, both of dual-image hiding scheme and CFS cannot reduce the frequent occurrence of the maximum absolute value, and this disadvantage diminishes the visual quality of the resulting stego-image. In the dissertation, three encoding methods are proposed to effectively decrease the modification level in the data embedding stage. One of them is the frequency-based encoding method that decreases the occurrence of the larger absolute digit and the level of modifications. The PSNR value of the frequency-based encoding method increases significantly in comparison with the previous methods. For smooth secret images, the image quality of the scheme is higher than those obtained from other methods. The second scheme is the similarity-based encoding method that integrates with the neighboring pixel similarity and the frequency-based encoding method, and then by utilizing the similarity of adjacent pixels of the codebook as the smaller coding value, the method further enhances the quality of stego-images. Experimental results show that the PSNR value of the similarity-based method in most images is higher than that of other similar studies. The third method is a dynamic encoding strategy. The strategy not only reduces the frequency of occurrence of the largest value but processes all the procedures of data encoding and transmits them immediately. With the advantage of the dynamic encoding technique, the proposed strategy has wide applications. In the future, the correlation coefficient among the embedding algorithm in the complex images will be analyzed to further enhance the quality of the embedded images. In addition, the proposed method integrated with the watermarking scheme will be studied to achieve a higher degree of robustness.