This is an idea for OTP (One Time Pad) for images. It is a couple of variations on an idea that is not mine. Perhaps it appeared in Scientific American a few decades ago. First the original idea:
In this scheme a transparency serves as the OTP. It is divided into pixel shaped areas that are randomly opaque or transparent. The pixel at (i, j), Pij = 0 for transparent or 1 for opaque. The OTP is indeed just the size of the image to be protected. The image is available to the sender in monochrome—actually one real number, gij, representing a point on the gray scale for each pixel in the image. The sender has a computer and produces for transport the cypher-image, a physical image whose pixel at (i, j) is pij which is 0 or 1 (transparent or opaque). pij = (gij<r) XOR Pij. The r above is a random number that ranges over the gamut of values of gray. A different r is taken at each pixel and need not be of cryptographic quality.
The recipient overlays the OTP and the cypher-image, aligns carefully, and sees a somewhat degraded plain image. If the recipient can perform an XOR instead of an AND then the image he sees is improved but still somewhat degraded.
This scheme provides secrecy as well as authentication. A fake cipher-image can produce noise at best, and noise that the opponent can not judge without access to the OTP.
If an opponent has write access to the cypher-image in transport, then he can further degrade the image but not substitute an image of his choice. The attacker can photographically reverse chosen parts of the image. He can put a black border on an image if he correctly guesses that the plain-image is light there.