Abstract: An information hiding algorithm is presented, which is based on coefficients relationship and fractal dimension The original image is divided into blocks with fixed size The relationship between fractal dimension and image blocks is used to determine the characters of each block Combined with characteristics of human vision system （HVS）, the blocks that represent the edges and texture are selected and DCT is applied to them Information is embedded into the relationship between DCT-coefficients Information includes character sequences, number sequences, text files, binary images, gray images and color images and so on Experimental results show that this algorithm can get good perceptual invisibility, high safety, and robustness against geometry attack, filtering, noises and compression