Hiding Data Reversibly in an Image Using Histogram of Differences between Pixels and Predictive Pixels

• Main Authors: Chiuan-Yu Liao, 廖全譽
• Other Authors: Chih-Ying Chen
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/38681239844530400696

碩士 === 逢甲大學 === 通訊工程所 === 98 === Image steganography is a secret communication technique that differs from the encryption / decryption methods used in cryptography. This technique is mainly achieved by using a meaningful yet unimportant digital image to cover and transmit the secret messages, thus denying any third party (or attacker) other than the sender and receiver from discovering these messages through the human eye or statistical detection. In 2007, Lin and Hsueh extended Ni et al.’s histogram reversible data hiding which used pixel grayscale values, and proposed a method that draws a histogram based on the absolute difference values from two neighboring pixels. Since neighboring pixel values usually have very little difference, most of these differences are highly concentrated in between 0 and 5, thus the peak of the histogram is also much higher than that of the original pixel values. Therefore, this method has a high embedding capacity of the secret message bits. In this paper, we apply the JPEG-LS predictive encoding to predict the pixel values. We compute the absolute difference between the original pixel values and the predicted values, which is then drawn in a histogram. Since the prediction method of JPEG-LS refers to the difference of neighboring pixels when determining the each estimated pixel value to decide whether an estimated pixel value belong to a horizontal, vertical edge or other cases. Therefore, the difference of the estimation and the actual value is quite small, generally ranging in between 0 and 3. Thus the peak in the diagram of our method is much higher than Lin and Hsueh’s approach, meaning that our embedding capacity is larger. Moreover, due to the fact that there is 50% chance of each embedded bit being 0, even under higher peak situation, we have more pixels that remained unmodified. Therefore, the results of our PSNR value (Peak Signal to Noise Ratio) used to determine the quality of a stego-image is superior than Lin and Hsueh’s value. In addition, we also modified the contents of the overhead information mentioned in Lin’s method, reducing the amount of extra information required to be recorded and thus raising the embedding capacity actually used to embed secret messages. Finally, through experiments, it has been proven that our method is also able to resist the RS statistical stegoanalysis proposed by Fridrich et al. in 2001.