| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系 === 103 === Abstract
This thesis proposes three data hiding algorithm; namely, reversible data hiding using dynamic adjustable histogram shifting algorithm (DAHS); a data hiding algorithm with the optimal capacity based on exploiting modification direction (OPEMD); a universal exploiting modification direction algorithm (UNIEMD)。
The first algorithm we present is DAHS algorithm, which achieves reversibility using a histogram-shifting approach.
The message embedding in our algorithm can be divided into three stages. In the first phase, according to parameters input by the user, an embedding location is selected as the ratio of 75-85% of histogram peak to decrease image distortion caused by the hidden message. The second phase selects the ratio of 65-75% of histogram peak as the embedding point to continue the multi-level reversible message embedding. This aims to increase the capacity by allowing more levels of reversible message embedding. The third phase deals with the overflow or underflow problem. In particular, we reselect the opposite direction of the zero point in the histogram and conceal secret messages using the original parameters given by users to further increase the capacity. Experimental results show that our two-stage approach can effectively reduce the distortion of the image and increase the embedding capacity. In comparison with our competitors, our DAHS algorithm provides more embedding levels, thus increasing the total embedding capacity. Our proposed algorithm offers considerable flexibility, allowing the user to reassign new parameters until a desirable capacity is produced.
We propose the second algorithm, OPEMD, which is an extension of the conventional EMD algorithm. Our algorithm inputs a single parameter, n, representing how many pixels are combined together to conceal a secret digit in the (2n+1)-ary notational system. We develop a universal encoding technique which uses large number operations to effectively convert secret bits into (2n+1)-ary digits. In our implementation, we employ relevant operation functions provided in the GNU multiple precision arithmetic library (GMP). Our OPEMD algorithm offers three advantages. First, our message conversion is so effective that the maximal bit loss due to conversion is less than one bit, achieving nearly optimal capacity. Our algorithm requires a single parameter, thus d to establish database to get optimal capacity. Finally, the utilization the universal encoding scheme avoids the bias of message conversion so the produced digits are more uniformly distributed, significantly reducing the risk of steganalytic attack from visualizing histogram of a stego image. We recommend the third algorithm, a universal exploiting modification direction algorithm (UNIEMD). Given a parameter n, we group n pixels into a number of sub-groups automatically to maximize the embedding capacity. Once the number of pixels in each sub-group is determined, we apply our OPEMD algorithm to embed secret messages providing the maximal and optimal embedding capacity. Our UNIEMD algorithm provides two benefits. First, our algorithm offers the optimal and maximal data hiding at the same time, increasing the embedding capacity substantially. Second, the pixel grouping is conducted automatically which not only maximizes the capacity but also increases the difficulty for eavesdroppers to unveil information for possible steganalytic attack.
In conclusion, we introduce three data hiding algorithms. The first algorithm provides higher embedding capacity than our counterparts. The second algorithm provides nearly optimal capacity, thanks to the universal encoding scheme we recommend. Finally, given a single parameter, the third algorithm automatically group pixels thus, offering both optimal and high embedding capacity. Our algorithms are feasible to data hiding applications.
|
|---|
