An 8-bit (1 byte) representation with the conventional integer to binary conversion. It is clear that choosing the right index for embedding is very crucial. This intricacy is less severe when using the RGC since it produces seemingly disordered decimal-to-binary representation. 

Context in source publication

Context 1

... algorithm using High and Low pass filters. In our case we compute four filters associated with the orthogonal or bi-orthogonal of the Haar wavelet. We choose Wavelet over DCT (Discrete Cosine Transform) because [3]: the Wavelet transform understands the Human Vision System (HVS) more closely than does DCT; Visual artefacts introduced by wavelet coded images are less evident compared to DCT because the wavelet transform does not decompose the image into blocks for processing. DFT (Discrete Fourier Transform) and DCT are full frame transforms. Hence any change in the transform coefficients affects the entire image except if DCT is implemented using a block based approach. However DWT has spatial frequency locality, which means if the signal is embedded it will affect the image locally. Hence a wavelet transform provides both frequency and spatial description for an image. More helpful to information hiding, the wavelet transform clearly separates high-frequency and low- frequency information on a pixel-by-pixel basis [4]. We refer to data that we wish to embed as payload, herein the image itself. Since we need means for protecting scanned documents against forgery it is essential that the payload will carry as much information from the host (cover image) as possible. There is a trade-off between perceptual visualization and space demand for embedding (usually measured in bits). Without taking compression into account, the payload can be consistent with the cover signal; therefore, if the cover is stored as an 8-bit unsigned integer type then the payload will require 8 templates when applying the one bit substitution method. There is a high payload Steganography approach called A Block Complexity Data Embedding (ABCDE) [5], but it is prone to statistical attacks as it acts in the spatial domain; moreover it cannot resist any kind of manipulation to the Stego-image (image having embedding data). An approximation of the cover document can be achieved through applying the gray threshold technique which results in a binary image demanding only 1 bit per pixel for storage. Some authors suggest using an edged image instead as it approximates the cover better. In the search for the best way to represent the cover image with the least bit requirement for embedding we identified dithering as our ultimate pre-processing step which is the foremost task in building our system. Dithering is a process by which a digital image with a finite number of gray levels is made to appear as a continuous-tone image [6]. Despite, in all versions, each pixel takes on only one bit, it is apparent that the way dithering quantizes image pixels contributes a lot to the final quality of data approximation. We observed that thresholding performs better in text based documents, while in capturing graphics it is proven to be a poor performer compared to dithering. Therefore, since our aim is to produce a general workable prototype we have to take into consideration the presence of both text and graphics; subsequently we opt to use dithering. Manipulating coefficients in the wavelet domain tends to be less sensitive unlike other transformations such as DCT and FFT. There are two methods to convert decimal integer to a binary string: one is to use the conventional decimal to binary conversion and the other is termed The Binary Reflected Gray Code (BRGC) 1 . This binary mapping is the key to the augmented embedding capacity introduced by ABCDE algorithm. There is a trade-off between robustness and distortion which is summarized in Figure 1. A trade-off occurs during our algorithm’s formulation which is due to the different levels wavelets can have. The lower we go the more robust we get but with less capacity for embedding. For example if the cover image is of size 255x255 (8-bit grayscale) we obtain 16384 bits to embed in the first level, the second level will reduce the image dimensions by a factor of two to yield 4096 bits and so on. In some cases the inverse transform in the wavelet domain truncates some values that fall larger or lower than the allowed limits in 8-bit type of images, the truncation occurs because of the introduction of “non-natural bits” coming from the secret message while embedding. To cope with this rare problem we choose to transform the RGB image into the YCbCr colour space prior to feeding it into DWT where we embed in the chrominance red channel (Cr). This step ensures that there will not be any data lost. Our proposed design is illustrated in Figure ...