The direction of incident light on an object is a strong physics-based cue for validating an image. Johnson and Farid proposed in 2007 a first approach to estimate the lighting environment in 2-D in the image plane. A user selects object contours along which the lighting environment is estimated. One limitation of this early approach is the requirement to select all contours along the same material. In our work, we relaxed this requirement by incorporating a material normalization term into the computation. That way, it is possible to mix, for example, contours of various garments and skin.

The paper can be downloaded here.

We also captured two datasets, a detailled description of the datasets can be found in the paper. The incandescent light dataset can be downloaded here, the flash dataset can be downloaded here.

If you use the data, please cite our paper
C. Riess, M. Unberath, F. Naderi, S. Pfaller, M. Stamminger, E. Angelopoulou: "Handling Multiple Materials for Exposure of Digital Forgeries using 2-D Lighting Environments", Multimedia Tools and Applications, 2016, doi:10.1007/s11042-016-3655-0 .