A Predictive SMVQ Steganographic Method Using Multiple Classification Codebooks

• Main Authors: Shu-Hua Lai, 賴淑華
• Other Authors: Chin-Feng Lee
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/49748199975056483105

碩士 === 朝陽科技大學 === 資訊管理系碩士班 === 96 === Due to the rapid development of the Internet and multimedia techniques, data hiding in digital media has received increased attention. Considerable quantities of researchers have devoted themselves to the study of watermarking and data embedding. Watermarking protects the copyright of multimedia products, while data embedding securely delivers invisible secret messages that are hidden in multimedia. The latter technology is generally referred to as steganography. Recently, some research was been proposed that integrates data hiding with image compression techniques such as vector quantization (VQ) and side-match vector quantization (SMVQ). Mean gray level embedding (MGLE) hides secret data in VQ compressed code, but the embedding capacity is poor. In 2003, Du and Hsu improved the capacity and visual quality of the cover image, but their scheme requires rearranging and regrouping the codewords according to the contents of the secret data. To improve Du and Hsu’s scheme, Shie et al. in 2006 proposed an adaptive data hiding scheme. They embed secret data into sufficient smooth blocks, keeping the SMVQ compressed code the same as the secret bits. However, since human eyes are quite sensitive to the sufficient smooth image blocks, changes to these blocks will bring attention to interceptors. In addition, considering imperceptibility, the embedding capacity will be confined within certain limits. Therefore, the embedding capacity and the image quality of Shie et al.’s scheme might have more space to be improved. For these reasons, we propose an adaptive data hiding scheme based on SMVQ prediction to improve Shie et al.’s scheme. PSMVQ edge detection technique is used to mask the edge direction of block, and then each block is classified accordingly. The image block identified by PSMVQ edge detection can be referred to be as the “edge block.” Accordingly, four codebooks are trained for those edge blocks. Just like Shie et al.’s scheme, a smooth codebook needs to be generated for encoding those smooth blocks. Contrary to Shie et al.’s scheme, secret data is embedded into those edge blocks and non-sufficient smooth blocks as well. In addition, the Edge-directed Prediction, also called EDP prediction for short, is applied to increase image quality. Similar to SMVQ prediction, EDP method exploits more neighboring pixel information to generate a set of state codebook whose codewords contribute to more accurate estimation of predicted block.