Image index tables values generally give the best possible representation of the color information of the image. However, no consideration is given to the arrangement of the color table itself. Thus, depending on the image, pixels with similar colors may have different index values and can therefore have considerably different index binary makeups. Consequently, regions of similarly colored indexed pixels can be noise-like at the bitplane level while the output colors themselves may imply simple bitplane patterns. BPCS image steganography hides information in images based on the principle that if regions in a bitplane are noise-like, those regions can be replaced with noise-like secret data. Therefore, applying traditional BPCS steganography to indexed image data results in drastic visible changes to the image. To overcome this problem, we used a self-organizing neural network to reorder the index table, based on samples from the image, such that similar colors in the index table are near each other with respect to their index values. As a result, regions with similar color information have only slight binary differences at the bitplane level, whereas regions with mixed color information will have considerable binary differences. Using this technique, we can embed secret data that is 15 to 35 percent the size of the image with little or no noticeable degradation in the image.
展开▼
