Method and device are described for customizing the content selection to present on a display device based on identifying the viewer of the device. In one embodiment, the present disclosure relates to selecting from among a group of digital images, those images that are most likely to be of interest to the viewer. In general, many of the images available to the display device will be comprised of images containing subject faces. Using the relationship information to predict subject affinity, the display device computes an image affinity for each image by accumulating the individual subject affinity predictions between the viewer and each subject identified in each image. The image affinities are used to select images for presentation on the display device.