US Patent 8218908 Mixed content image compression with two edge data representations

A method (200, 400) of compressing an input image (201) is described. The method (200, 400) starts by extracting image gradient data representing pixel differences along contours of the image. Next, the method forms an edge based image from the image gradient data and a low resolution image derived from the input image. The method then computes a residual image representing a difference between the edge based image and the input image, and finally compresses the input image by encoding the image gradient data, the low resolution image, and the residual image.