Specular Reflection Image Enhancement

We've all seen Rick Deckard and apparently the people on CSI enhancing images to a ludicrous extent, enabling the image to reveal objects on the other sides of walls and under the floorboards or whatever. All of this is clearly ludicrous. At the same time I still think my "gainy image compression" (see link) would work to some extent. I also think this would.

I was recently perusing a picture of myself from a house I lived in a few years ago and noted that because I happened to know a specular reflective surface, to wit a metallic zip fastener, was facing a scene including a carpet, a piano stool, upright piano and street scene through a window, I was able to glean which teeth were reflecting which objects. At the same time the lenses on a pair of glasses reflected a much more easily enhanced image of the same scene.

Just now, I went into the kitchen to note the presence of a chrome milk jug, kettle, cutlery, a wine glass, metallic draining board and sink reflected from the aforesaid chrome jug, and was able to infer from those reflections and refractions while facing away from the window that there was a window with a pair of curtains, some kind of garden and a partly clouded sky.

Therefore, it seems to me, and don't ask me how, that if something like raytracing could be performed in reverse in order to reconstruct scenes which are not obviously in the image and away from which the camera is facing. I can vaguely suggest how but there would need to be a mechanisable bit which I have little idea how to approach.

It amounts to computer vision and then some. The specular reflections in an image clearly depend on the angle of the surfaces from which they reflect. If there's a way of identifying areas of reflection and specular reflection and calculating those, the likes of refraction through transparent media and off glass, metal and smooth surfaces of other kinds could be used to reconstruct the unseen parts of a photographic image. The same applies to video, but in this case the job may be easier because movement would give clues as to the shapes and angles concerned.

I realise this is a big vague blob of an idea with a few little details worked out round the edges but surely this could be done one day, couldn't it?