In an image processing device, an edge image generation part detects an edge in an original image and generates an edge image constituted from the detected edge. A connected pixel extraction part extracts connected pixel groups in the edge image. A binary image generation part classifies the connected pixel groups under respective colors of the connected pixel groups, and generates a character image for each color. A background image generation part generates a background image of the original image based on the character image so that a pixel value at the position of the character image in the original image is set by an average value of the pixel values in the original image with regard to at least a portion of pixels around a rectangle circumscribing the connected pixel groups. An image compression part compresses respective image data of the character image and background image by different compression manners.