Have you ever looked at a picture and wondered if it had been digitally edited? Now, you may finally find out.

Micah Kimo Johnson of the cognitive sciences department of MIT created a program to detect fake pictures. The theory behind his method is that fake pictures are usually made by copying and pasting from different photos that have different lighting sources. This reminded me of the shadowbox because the purpose was to detect the light source of an image.

Explained in general terms, the program generates a lighting source matrix from the picture. Then the user selects a subimage, and calculates the lighting matrix for that. If the two values fall beyond a certain threshold of similarity, then the picture is declared a fake. However, you need to know the general 3d shape of the subimage you want to check.

In the paper, Johnson makes some key assumptions in order to generate a lighting matrix. With this assumption he creates the following equation. Intensity anywhere on the picture is a function of reflectance, the normal vector times the lighting vector, plus the ambient lighting level.

Johnson solves for L using the least squares estimate. The full details of the algorithm can be found in a link to the research paper below.

Example from original research paper.

You can see that fake images are a big problem which scientific computing can solve.

http://www.mit.edu/~kimo/research/digital_forensics/lighting.html