With my hero Alan Turing's centenary coming next year, I thought he would make an ideal subject for an experimental 3D-printed lithophane http://en.wikipedia.org/wiki/Lithophane.
While Litho-Alan makes a wonderful window decal, he is really just a proof-of-concept that you could add to things like lampshades and candle holders. The nice thing about lithophanes is that you only need a black-and-white picture as the source, not a full 3D model.
Craftsmen and artisans of the pre-industrial era would often personalize their work, perhaps with a picture of the patron who commissioned the piece. With 3D printing we are able to make just one of something and to tailor it specifically to one person.
The heavy lifting for this was done by Sven Geisel's brilliant bitmap-to-STL conversion tool http://www.ba-bautzen.de/~geisel/bmp2iges.htm. It basically converts a grayscale image into a height map, which you feed into skeinforge. I've included a screenshot of my conversion settings, but you should experiment to find the best fit for your medium and printer.
I used 1.75mm white PLA (from UltiMachine) with a 0.3mm layer height (0.15mm on the first layer I think). Other plastics will have different opacities, so you may get better contrast by tweaking the settings. The print is 100mm x 130mm, which is right to the edge of my print bed. Don't forget to set the in-fill to 100%.
The good news is that the BMP2STL conversion takes seconds, the bad news is that skeinforge took six hours to render it to G-code! The 32-bit version choked above 1000x1000 pixels, but that is beyond the limit of printability anyway so I dropped the resolution. To be fair the STL file is a mesh of thousands of tiny polygons, which is not straight forward to translate. I've uploaded my PLA based G-code in case it helps, but it probably contains assumptions specific to my printer (an eMaker Huxley). The crazy G-code skeinforge creates will put a strain on your stepper motors for which I take no responsibility. The print itself took several hours, but used only a modest amount of plastic (roughly 20m).
Clearly an alternative conversion process would help. The biggest strain comes from the high polygon count in the (121MB!) STL file, which is itself a product of the complex textures in the photograph. I started looking into piggy-backing some of the bitmap-to-polygon conversion functions in Inkscape and thence to G-code, but didn't have the time to learn Python. At the very least, some pre-processing of the photograph to remove unresolvable artefacts helps and I did boost the contrast a bit to completely saturate the background. I've included the original source image and the cleaned up image (BMP file) I rendered the STL from.
If you do any successful conversions please hit the comments section with hints-and-tips!