> On Fri, Aug 14, 1998 at 10:59:44PM +0200, Joost Witteveen wrote: > > Hi, > > > > I've got (at least) one photo taken while the camera was moving. > > Only one? :)
Well, one that I care about. Much more, actually. > > > I'd really like to get rid of the vageness caused by this motion. > > Mmmh. I doubt that this is possible... but I'd like to be proofed wrong. > Actually, missing sharpness is lost information. Actaully, it isn't. If you take two pictures, and superimpose both of them slightly (or more) displaced, it _is_ possible to excactly calculate back what the original was (assuming you know what the displacement was). With a moved picuture it will probably be somewhat more difficult (you don't know the displacement, and it's not two superimposed pictures, but more like a continuas mess of them), but it should still be possible (just a lot more work). > How can this information be > regained? OK, one simple 1D example (a 2D picture is just excactly the same) assume the intensity of the `real' picture is 00220022200002400 (with 0 begin black, 4 being white). Now assume the camera was displaced one `step` halfway during the exposure time: 00121012210001320 will be the result. Now, with a few assumptions (the real difficulty is in the edges, but assume one knows the outer two positions were black), it is quite easy to calculate the image back: The dot on position 3 of the `moved' image being 1 proves that the real intensity was 2 (we know the first two positions are black, because otherwise the `moved' image can never have two initial black pixels, and so the `1' intensity was half of the real intensity of position 3). The `real' dot on position 3 begin intensity 2 means that half that intensity was moved over to dot position 4. So of the intensity `2' of that `moved pos 4' dot, 1 part is from dot 3, and 1 part must be from dot 4. Thus also dot 4 must be intensity `2'. And so on. > OTOH, maybe there are acceptable solutions for certain types of > photos. > > The only hint I can give you is reading the gimp manual. Maybe there are > some tricks in there. Only the `sharpen' stuff. What I want is a function that atually calculates the displacement (not possible in the general case, but nearly always possible in reality), and then inverts the `motion' joost witteveen
