Abstract: An algorithm for segmenting document image‚taking into consideration both noise of background and interference of texture is proposed to improve the compression ratio for scanned document image.First the multiscale characteristic image with reduced texture of text is generated.Then picture areas in document are extracted by means of marked segmentation based on connection degree of pixels.At last picture and text areas are demarcated by fusion of different scale characteristic images. Experiments show that this algorithm is simple and effective.