Go to | Data for computer vision | Kobus Barnard 's research page | Kobus Barnard 's Home Page

Test Images for Computational Colour Constancy

The official source for this data is elsewhere, and the the appropriate web resource reference to the data is one of
However, the associated meta data is likely to be corrected/updated more frequently in this version of the interface.

Questions, comments, and problems with this data should be directed to Kobus Barnard

This directory contains some the data presented in:

Kobus Barnard, Lindsay Martin, Brian Funt, and Adam Coath, " Data for Colour Research," Color Research and Application, Volume 27, Issue 3, pp. 148-152, 2002.

(The appropriate archival reference for this data).

This data in this directory is used in the following publications:

Kobus Barnard, Brian Funt, Linday Martin, and Adam Coath, " A comparison of color constancy algorithms. Part two. Experients with image data," IEEE Transactions on Image Processing, Vol. 11, No. 9, pp. 985-996, 2002.

Kobus Barnard, " Color Constancy with Fluorescent and Surfaces," in preparation.

Kobus Barnard and Brian Funt, " Color Constancy with Specular and Non-Specular Surfaces," In preparation.

Kobus Barnard, Brian Funt, and Lindsay Martin, " Color Constancy Meets Color Indexing," unpublished manuscript, 2000.

Kobus Barnard, "Practical Colour Constancy," Phd thesis, Simon Fraser University, School of Computing (1999)

Kobus Barnard and Brian Funt, " Color Constancy with Specular and Non-Specular Surfaces," Proceedings of the IS&T/SID Seventh Color Imaging Conference: Color Science, Systems and Applications, 1999, pp. 114-119.

Kobus Barnard, "Color Constancy with Fluorescent and Surfaces," Proceedings of the IS&T/SID Seventh Color Imaging Conference: Color Science, Systems and Applications, 1999, pp. 257-261.

The data was collected by Lindsay Martin, Kobus Barnard, and Adam Coath, under the guidance of Kobus Barnard in Brian Funt's Computational Colour Vision Laboratory.

Data Description

The data consists of a number of scenes taken under 11 different lights. These lights were chosen to be representative of the spread of common illuminants. We divide the scenes into four sets. These are a set of images with minimal specularities (21 scenes), a set with non-negligible dielectric specularities (9 scenes), a set with metallic specularities (14 scenes), and a set with at least one fluorescent surface (6 scenes). Some images were culled due to deficiencies in the calibration data. There remained 223 valid images in the first set, 98 in the second, 149 in the third, and 59 in the fourth.

The experimental routine was as follows: First a new scene was constructed. We then placed a reference white standard in the center of the scene, perpendicular to the direction of the illuminant. The position of the illuminant was set so that the number of clipped pixels was small. This meant that if the scene had bright specularities, then the image was purposely under-exposed. We then took a picture of the scene with the reference white in the center, and captured the spectra of the light reflected from the reference white. Finally, we removed the reference white, and took 50 successive pictures which were averaged to obtain the final input image. We then repeated the process for the remaining 10 illuminants, and then we moved onto the next scene.

The images with the reference white were used to provide the answer. We extracted the central 30 by 30 pixel window of each of these images, and used the average (R,G,B) over these windows as the estimate of the illuminant for the corresponding input images. We note that both the input image and this target value were first mapped into a more linear space, and received the other corrections discussed more fully below. We believe that this method provided a good estimate of the chromaticity of the illuminant, but that the error in the illuminant magnitude for any given picture could be quite high­easily 10%, because of the difficulties in keeping the white reflectance standard perpendicular to the light source. Furthermore, three of the sources were distended, and here we simply attempted to find the orientation which maximized the brightness of the reflectance standard.

Because of the frame averaging described above, some of the images have very large dynamic range, and the image pixels have more than the usual 8 bit precision which was preserved by performing the averaging with floating point arithmetic and storing the results in a floating point format. Several pre-processing steps were taken to improve the data. First, we removed some fixed pattern noise. Second, we corrected for a spatially varying chromaticity shift due to the camera optics. Finally, we mapped the images into a more linear space as described in [ 1 ]. This included removing the sizable camera black. The resulting images are such that pixel intensity is essentially proportional to scene radiance.

Because of the preprocessing and extended dynamic range, it is possible to scale the images by a factor of up to about 10 without incurring too much noise. Thus the images can be re-scaled to emulate capture with an automatic aperture. For example, if an scene has significant specularities, then the specularities would normally be clipped. We tried to minimize clipping in our images, but normal camera behavior can be emulated by scaling the image up, and clipping the result at 255, or whatever level is appropriate for the research being done. Thus our approach allows the study of higher dynamic range images, which may become more readily available and are worthy of consideration in many applications, but does not rule out the emulation of more standard camera behavior.

In order to allow researchers to experiment with the extra data depth we provide the images as 16 bit tiff images, as well as 8 bit tiff images, which may be more convenient for many people. There are links below to each of the four sets of images. Each set of images is provided as a gzipped tar file. There are two such files for each set, one for the 8 bit tiff images, and one for the 16 bit tiff images. Each unpacks to a directory of the form <set>_<n>_bit, where <set> is one of "mondrian", "specular", "metallic", and "fluorescent", and <n> is one of 8 or 16. These directories have sub-directories for each scene. In each sub-directory there are up to 33 files (less if the scene had some images culled). In the non-culled case, these 33 files consist of 3 per scene, one for the image, one for the illuminant RGB, and one for the illuminant spectra. Because of imperfections in the calibration, we suggest that the illuminant RGB is used where possible.

1. See any of the references listed on this page.

Links to Each of the Four Data Sets

Images with few specularities (231 images)

Images with some specularities (98 images)

Images with metallic specularities (147 images)

Images with fluorescent surfaces (57 images)