Database of Contaminants' Patterns

Textures or patterns of the contaminants play an important role for photorealistically reproducing the effects. We model the patterns as 2D optical thickness texture, τ(x,y). To measure τ(x,y), we use the shadow map generated by attenuation from the contaminant layer. The following image shows our setup. The projector illuminates a thin glass slab with contaminants on the far side. Behind it is a Lambertian board, and the camera is on the side. The camera is radiometrically calibrated beforehand. The intensity of each point in the shadow map comes from two parts: the attenuated transmitted light from this point, and the scattering component from neighboring points. Since the albedo of the contaminant layer is assumed to be small (otherwise it would generate multiple scattering) and it is mostly forward scattering, there will be much less contribution from neighboring points due to scattering. Thus the attenuated light is much stronger than the scattered light and it is the intensity of the shadow map. For some samples we can indeed observe the effect of scattering from neighboring points in the shadow map, for which the shadow map usually is blurry, especially when the board is far from the glass. In those cases, we let the projector shine light through the glass and put the camera on the other side. This setup allows us to measure τ(x,y) by scattering and the result is then scaled by the initial measurement.


The database is the property of Columbia University. This data can only be used for research or academic purposes. Any commercial use of the data whatsoever or incorporation of the data into a larger database intended for public distribution must be done with the explicit written consent of CAVE administrators.


The database contains 36 samples, including various kinds of contaminations, such as dust, dirt, fingerprints, lipids, soap water deposit, oily smudges, clay, and so on. The image intensity is propotional to the optical thickness. We assume there is only one kind of contamination for each sample. The images are available in two formats: EXR and PNG. EXR is a format for high dynamic range images. Here we assume the contamination layer is very thin, so the image dynamic rang is in fact not very high. But still, floating point can give us more accurate resolution. For more information about EXR, visit OpenEXR please.

Click each of the sample in the following table to download. Also, the 36 samples are zipped and can be downloaded together in EXR (163MB) and PNG (17MB). We provide some Matlab code for reading and writing EXR files, as well as a simple synthesis tool which allows the use to interactively generate new large textures from the acquired samples here.

Database Home
Last modified: 06/02/2007