Depth constrained superpixel-based depth map refinement

A method of forming a refined depth map DR of an image I using a binary depth map DI of the image, said method comprising segmenting (315) the image into a superpixel image SREP, defining (330) a foreground and a background in the superpixel image SREP, to form a superpixel depth map DS, intersecting (450) the respective foreground and the background of the superpixel depth map DS with the binary depth map DI determined independently of the superpixel image SREP, to define a trimap T consisting of a foreground region, a background region and an unknown region, and forming the refined binary depth map DR of the image from the trimap T by reclassifying (355, 365) the pixels in the unknown region as either foreground or background based on a comparison (510) of the pixel values in the unknown region with pixel values in at least one of the other trimap regions.