An image processing unit comprises a first processing unit (101) which generates a depth indication map for an image. The depth indication map may be, for example, an image object separation mask or a predetermined depth profile or background depth map. A second processing unit (103) generates a modified depth indication map by filtering the depth indication map in response to image characteristics of the image. The image adaptive filtering may, for example, provide a more accurate separation mask and/or may modify the predetermined depth profile to reflect the specific image. A third processing unit (105) generates an image depth map for the image in response to the modified depth indication map. The image depth map comprises data representing a depth of at least one image region of the image. The invention leads to the generation of an improved depth map for an image.