A method for automatically producing a new digital image from a first digital image that includes regions of interest, includes obtaining a main subject belief map including an array of belief values indicating the location and relative importance of subject matter in the first digital image; producing a mask for each privileged object in the first digital image, each such mask including margins around its corresponding privileged object; overlaying the mask(s) onto the belief map; and producing a first convex hull that includes the mask(s) with margin(s) and regions of the highest belief values from the belief map.