01-01-2016, 04:31 PM
Abstract
Most state-of-the-art binary image steganographic techniques only consider the flipping distortion according to the human visual system, which will be not secure when they are attacked by steganalyzers. In this paper, a binary image stegano- graphic scheme that aims to minimize the embedding distortion on the texture is presented. We extract the complement, rota- tion, and mirroring-invariant local texture patterns (crmiLTPs) from the binary image first. The weighted sum of crmiLTP changes when flipping one pixel is then employed to measure the flipping distortion corresponding to that pixel. By testing on both simple binary images and the constructed image data set, we show that the proposed measurement can well describe the distortions on both visual quality and statistics. Based on the proposed measurement, a practical steganographic scheme is developed. The steganographic scheme generates the cover vector by dividing the scrambled image into superpixels. Thereafter, the syndrome-trellis code is employed to minimize the designed embedding distortion. Experimental results have demonstrated that the proposed steganographic scheme can achieve statistical security without degrading the image quality or the embedding capacity.
INTRODUCTION
TEGANOGRAPHY, analogous to the data hiding, aims to hide secret information under digital media in such a way that no one, apart from the sender and receiver, can detect the existence of the information. In recent years, many data hiding methods have been developed for binary images , which can be used to authenticate digitally stored hand- writings, CAD graphs, signatures, and so on. Stego images obtained by these schemes have also been reported to achieve considerable visual qualities. However, these methods ignore the security against steganalyzers. The high undetectability of the secret messages can reduce the suspicion from attackers and thus enhance the security. To this end, we focus on designing a secure binary image data hiding scheme (or more strictly speaking, a steganographic scheme) by improving the undetectability while preserving the stego image quality and embedding capacity.In the spatial domain, message bits are commonly embed- ded by directly flipping pixel values in a binary image. Unlike grayscale images, pixels in binary images possess only two states: black (1) and white (0). As a result, distortions on binary images are easily detected even by human eyes. To deal with this problem, practical steganographic schemes suggest constraining the embedding to the portions of images that are difficult to be noticed.
Some schemes traced the boundary to find more suitable pixels for embedding message bits , whereas the others divided the cover image into overlapped/non-overlapped blocks and found the best flipping location in each block .By employing 2 ×2 size blocks and double processing, the scheme presented in used nearly all the shifted edges to embed message bits and thus achieved a large payload. Matrix embedding is usually employed to achieve a high embedding efficiency. Filler et al. [8] proposed a practical near optimal matrix embedding, namely syndrome-trellis code (STC), to embed near the capacity- distortion bound with respect to the specified distortion measurement. Prior works also supported the priority of STC . Consequently, we employ this code to implement our steganographic scheme. The above-mentioned schemes all measure the embedding distortion according to the human visual system (HVS). Therefore, the yielded stego images present good visual qualities and usually cannot be distinguished from the cover images by human eyes. However, we know that the adversary may reveal the secrets with the assistance of steganalyzers. As reported in Section IV-C, these schemes seem to be insecure in this case. To make a steganographic scheme secure, an advantage way is to model the image statistic and minimize the embedding impact on that model . Noting that binary images naturally represent the texture, we exploit the texture model to measure the embedding distortion. Broadly speaking, there are three types of approaches describing the texture : geometry-based, statistic-based, and model-based approaches. In the proposed measurement, the first and second types are combined to describe the texture with respect to both spatial structure and statistical distribution. That is, we first extract the local texture pattern (LTP) as the texture primary. The histogram of LTPs is then employed to describe the texture distribution.
The LTP is motivated by the concept of the local binary pattern (LBP), which has been successfully applied in texture classification [16], face detection [18], steganalysis [19], and so on. Since binary images possess different visual appearance compared with grayscale images, an extension of the LBP, namely the complement, rotation, and mirroring-invariant local texture pattern (crmiLTP), is developed to be better applied in binary image steganography. We know that the texture region is more suitable for steganography [10], [20]. Therefore, it is expected that a good stego system can be achieved in virtue of the texture model.The distortion measurement needs to coincide with HVS and statistics simultaneously. Unlike the texture-based measurement proposed, there have been approaches handling distortions by employing the HVS [3], [4], [21], [22]. Among them, Wu and Liu [3] assessed the flipping distortion according to the smoothness and connectivity in a 3 × 3 window.Yang and Kot [4] defined a connectivity-preserving criterion for 3 × 3 patterns to determine the flippability. Lu et al. suggested using the distance reciprocal distortion measurement to measure the distortion effect on the neighboring pixels, and Cheng and Kot [22] presented an edge line distortion-based criterion to describe the distortion on the boundary connectivity.
In this paper, the proposed measurement is compared with them by using an ideal embedding simulator.In this paper, a spatial domain-based binary image stegano- graphic scheme is proposed. The scheme minimizes a novel flipping distortion measurement which considers both HVS and statistics. This measurement employs the weighted sum of crmiLTP changes to measure the flippability of a pixel. Further, the weight value corresponding to each crmiLTP is set according to that pattern’s sensitivity to the embedding distortion. To estimate the sensitivity, a collection of general- ized embedding simulators are organized to yield stego images with different distortion types and strengths. In the embedding phase, STC is employed to minimize the flipping distortion. To remove the unexpected flipping incurred by STC, the concepts of scrambling and superpixels are employed to guarantee that flippable elements occupy the majority in a cover vector. By incorporating the new distortion measurement with the STC framework, the proposed steganographic scheme presents a significant performance compared with state-of-the- art works.
The reminder of this paper is organized as follows. The crmiLTP and the flipping distortion measurement are devel- oped in Section II. In Section III, the proposed steganographic scheme is presented. Comparison experiments among different distortion measurements and among different steganographic schemes are reported in Section IV. Finally, Section V concludes the whole paper.