Error correction would be a PITA the pixel values would be highly dependent on lighting, camera sensor (even between identical sensors), lens aberration and focus and any post processing the camera does both in hardware and software.

You will most likely have to use only primary colors or something quite close to that which will limit your data density considerably.