Survey of Related Works

In this study, the concerned topics include Image authentication, covert communication, and data hiding. With the era of cloud computing coming, data stored originally in personal computers mostly will eventually be moved to and processed in powerful servers at far ends. A problem involved here is how to make sure that personal data accessed from cloud servers are intact. This problem of data security is a very significant issue. This study explores this issue of keeping digital image data secure, and proposes the use of an image authentication technique as a solution.

In the literature, several authentication methods for binary images have been proposed. Wu et al. [1] explored flippable image pixels in which data can be embedded without causing noticeable artifacts for the purpose of authentication and annotation of binary images. Yang et al. [2] proposed a pattern-based data hiding method for binary image authentication, which uses the flippability of pixels effectively by a connectivity-preserving criterion [3]. Kim et al. [4] chose a set of pseudo-random pixels which were later cleared for embedding authentication code in a binary or halftone image. Tzeng and Tsai [5] embedded authentication codes into

blocks of a cover binary image for the later use of tampering detection in the verification process. In addition, to reduce the image distortion caused by the authentication codes embedding, a technique of using a code holder allowing multiple locations of pixels for embedding the codes was proposed. Lee et al. [6] embedded watermark data into one pixel in each binary image block by a Hamming-code-based data embedding method with small distortions and low false negative rates. Method proposed in [7] utilized an edge-line similarity measure to choose flippable pixels, yielding smaller distortion compared with that of method in [6].

As to the image authentication for grayscale images, several fragile watermarking techniques for image authentication have been proposed in the past and they may be categorized into two approaches: block-wise [8-14] and pixel-wise [15-18]. Methods of the former approach embed fragile watermarks as authentication signals into non-overlapping blocks of the cover image and identify possible tampered image parts in the unit of block. One weakness of such block-level authentication methods is that the detail of the tampered image part cannot be located precisely [16].

On the other hand, methods of the second approach [15-18] authenticate images at the pixel level such that tampered image parts can be identified pixel by pixel, yielding a detailed tampering localization result. Liu et al. [15] generated a binary image that is mapped from the difference image computed from the cover image and its so-called chaotic pattern. And the least-significant-bit (LSB) plane was used to accommodate the binary image as the fragile watermark for the use in later image authentication.

Because of the binary nature of the embedded fragile watermark, the LSB of a tampered pixel value may coincide with the watermark bit, yielding a high erroneous pixel authentication rate up to 50%. To deal with this phenomenon, a statistical fragile watermarking method which utilizes probability distributions computed from the original pixels and the tampered ones to locate the tampered pixels was proposed in Zhang and Wang [16]. However, the method only works in the case that the tampering ratio is smaller than 1.1% [17]. As an improvement, Zhang and Wang [17] proposed later a fragile watermarking method for authenticating grayscale images using a hierarchical mechanism, which embeds watermark data derived from the pixels and blocks of the cover image into the LSBs of all the pixels. In the authentication process, tampered blocks are identified first, and tampered pixels within the identified blocks are located subsequently.

Covert communication is a technique of concealing secret information into a cover medium in an imperceptible way or with a camouflage effect such that only a sender and an intended receiver know the existence of the hidden data in the resulting stego-medium. In the literature, emphases were put on the use of multimedia like images, videos, and audios [19-22] because these media in general provide larger embeddable spaces and cause less suspicion due to their wide distributions. And weaknesses existing in human beings’ visual capabilities are often exploited to design effective covert communication methods. For example, the methods proposed in [23-25] replace the least-significant bits of pixels in cover images to embed information, and that of [26] uses the parities of palette colors, composed by similar colors, to represent hidden message bits. function in Microsoft Word to embed data imperceptibly by a document degeneration technique.

As to the topic of data hiding, many data hiding methods for use in the spatial domain of image files [31-35] have been proposed. Bender et al. [31] proposed the technique of least-significant-bit (LSB) replacement, in which a secret message is embedded in the least significant bits of image pixel values. The method yields high data hiding capacities and low computational complexities. Following Bender et al.

[31], Wang et al. [32] proposed an optimal LSB-replacement-based method which is integrated with a genetic algorithm to reduce the computation time for finding the optimal hiding result. Mielikainen [33] proposed a modified LSB replacement method which embeds as many bits as the conventional method, but changes less pixel values.

Wu and Tsai [34] proposed a steganographic method for images by pixel-value differencing, in which the difference values between pixel value pairs are classified into a number of ranges to decide the number of bits which can be embedded into the pixel pairs. Yang et al. [35] proposed an adaptive k-LSB substitution method in which larger values of k are adopted in the edge areas of the cover image and smaller ones are used for the smooth areas. Also, the range of value differences of consecutive pixel pairs is divided into different levels, and the value of k is adaptively decided

according to the level into which the pixel difference value falls. In this way, larger data hiding capacities with higher stego-image qualities can be obtained.

Another data hiding approach is to utilize the frequency domain. Wang et al. [36]

transformed image block contents into coefficients in the frequency domain by the discrete cosine transform (DCT), recomputed the AC values of the central block of every nine 8×8 image blocks, and embedded secret bits by modifying the magnitude relations between the new AC values and the original ones. Wu and Hsieh [37]

embedded data in the frequency domain by using the so-called zerotree in the rearranged DCT coefficients of the cover image. Besides data embedding techniques using the DCT, the discrete wavelet transform (DWT) [38-40] and the discrete Fourier transform (DFT) [41-42] have also been used.

From another viewpoint, different types of image may be used as cover media such as JPEG [43-45] and palette-based images [46-49]. Fridrich and Du [46]

explored the optimal parity of a color palette, and embedded message bits into the parities of palette colors. Tzeng, Yang, and Tsai [47] hid secret data into data-embeddable image pixels by using a color-mapping function, and the color of a data-embeddable pixel is replaced by an optimal one selected from the color palette.

Zhang et al. [48] utilized the concept of gregarious color, meaning a color with some other colors similar to it, in the palette to hide at least one secret bit for any pixel with a gregarious color. In [49], Lee and Wu proposed a reversible data hiding method for palette-based images, which adjusts palette colors and image data to embed secret data and side information for reconstruction of the original image content.

Reversible data hiding methods embed message data into host images and are capable of restoring resulting stego-images to their original states after the hidden data are extracted. Histogram shifting is an efficient technique used in many reversible data hiding methods. Ni, et al. [50] shifted slightly part of a histogram between its peak point and the zero point by one pixel value to create an empty bin besides the peak point for accommodating message data. The knowledge of the locations of the peak point and the zero point of the histogram need be memorized in order to retrieve the hidden data and restore the original image losslessly, making the resulting method non-blind. Kuo et al. [51], Lee and Tsai [52], and Fallahpour et al. [53] proposed a similar idea of decomposing an entire cover image into blocks and using the peak point of the histogram of each block to hide data. The technique of block division successfully improves the data hiding capacity but the side information composed of

peak points and zero points still should be kept. Tsai et al. [54] explored the similarity of neighboring pixels in the host image and employed a linear prediction technique to generate a residual histogram of the image for data embedding. The data hiding capacity of Tsai et al.’s method [54] is superior to the aforementioned histogram-based hiding methods and maintains simultaneously the high quality of the resulting stego-image; however, the side information of peak points and zero points is still needed. Later, Kim et al. [55] utilized the high spatial correlation between sub-sampled images of the cover image to obtain a high data embedding capacity while keeping the distortion in the stego-image at a low level.