Invariability of Mean Value Based Reversible Watermarking

By employing invariant relation between the mean value of the first (n - 1) pixels and the last one pixel (also called the remaining pixel) for every image block containing n pixels, a new reversible watermark scheme capable of mostly carrying 2n-3 bits into one n-sized image block in a single embedding process is presented in this paper. First, the mean value of the first (n - 1) pixel is calculated. Next, the difference value between the last one pixel and this obtained mean value is applied to distinguish which classification (i.e., smooth or complex sub-block) any sub-block belongs to. Consequently, it is determined to embed (n - 2) bits or 2(n - 2) bits into each sub-block according to its final classification results. And meanwhile, the mean value is reapplied to predict this last one pixel. 1-bit watermark is embedded into this last one pixel in accordance with the magnitude of prediction-error value. By multi-employing invariability of the mean value of (n-1) pixels, the embedding rate can approach to (2 - 3/n) bpp (bit per pixel) for a single embedding process. Meanwhile, the embedding distortion is greatly controlled by embedding more bits into smooth image blocks and fewer bits into the other blocks with complex texture. Experimental results reveal the proposed scheme is effective.