This is why image processing and computer vision excite me. The software can "understand" the image well enough to fill in unknown areas based on contextual information. I recall seeing SIGGRAPH papers on this sort of thing over the last decade, but there's clearly been a lot of progress.