|
|
|
Descriptions Information about these datasets Data Publications |
Synthetic Data for Computational Colour Constancy Experiments
This page contains links to some of the data described in:
(This is the appropriate archival reference to this data). Questions, comments, and problems with this data should be directed to Kobus Barnard.
Information on the Sony DXC-930 calibration data can be found on the camera calibration page. This calibration data was used to derive the camera sensors used for many of our synthetic data experiments. Important: Illuminants vary greatly in brightness. The spectral data provided here is the raw data reported by our spectrometer. For some applications, this data needs to be normalized. The first illuminant set (image_data_sources.illum) is a set of 11 sources chosen for most of our experiments with real images. The illuminant sources were: Sylvania 50MR16Q (12VDC)---A basic tungsten bulb The second illuminant set (measured_with_sources.illum) is a larger set of measured spectra. This set consists of 81 spectra measured in and around the SFU campus at various times of the day, and in a variety of weather conditions. Unusual lighting, such as that beside neon advertising lights, was excluded. However, care was taken to include some reflected light, provided that it was not too extreme. This set of illuminants was augmented with the measurements of 21 sources, including the 11 listed above. All these illuminant sources are plausible common illuminants. Possibly more convienient for most applications is the same set normalized to maximum of one. The third illuminant set (train.illum) was used for neural network training. To create this illuminant training set, we divided (r,g) space into cells 0.02 units wide, and placed the 11 illuminants described above into the appropriate cells. We then added illumination spectra from set of the measured spectra, provided that their chromaticity bins were not yet occupied. Finally, to obtain the desired density of coverage, we used random linear combinations of spectra from the two sets. This is justified because illumination is often the blending of light from two or more sources. In addition, to the extent that the diagonal model holds, these constructed illumination spectra will behave like physical sources with the same chromaticities as the constructed ones. The fourth illuminant set (test.illum) was used to test trained neural networks. To produced the testing illuminant set, we used the same procedure described for the training set, but filled the space 4 times more densely. Surface Reflectance Data The surface reflectance data (reflect_db.reflect) is a set of 1995 spectra compiled from several sources. These surfaces include the 24 Macbeth color checker patches, 1269 Munsell chips, 120 Dupont paint chips [1], 170 natural objects [1], the 350 surfaces in Krinov data set [2], and 57 additional surfaces measured by ourselves.
[1] M. J. Vrhel, R. Gershon, and L. S. Iwan, ?Measurement and Analysis of Object Reflectance Spectra,? COLOR Research and Application, vol. 19, pp. 4-9, 1994.
Camera Sensors: Illuminants Used With Image Data Experiments: Measured Illuminants: Neural Network Training Illuminants: Neural Network Testing Illuminants Reflectance Database:
|
