Fifty hyperspectral reflectance images of outdoor scenes

The 50 zipped files hsi_1.zip, ..., hsi_50.zip in this dataset contain averaged hyperspectral reflectance images of outdoors scenes in Matlab MAT format, each of size ~ 336 × 256 × 33. Each hyperspectral reflectance image was obtained by spectrally scaling an original radiance image 1024 × 1344 × 33 by the reflectance of one or more reference surfaces embedded in the scene. The reflectance image was spatially downsampled to ~336 × 256 pixels by averaging over 4×4 pixels and then smoothed spectrally by averaging over adjacent wavelength values 400, 410, ..., 720 nm.

The file Scenes_and_Names_Key.pdf contains a key for relating image label, e.g. hsi_1, to scene name and image appearance.

The three zipped illuminant files illum_4000.zip, illum_6500.zip, and illum_25000.zip contain daylight spectra in Matlab MAT format, each of size 33 × 1, drawn from CIE daylight spectral distributions, which can be used to test the effects of illumination changes on scenes.

The hyperspectral images were acquired by Sérgio Nascimento, Kinjiro Amano, and David Foster from the Minho region of Portugal, 2002-2003.

For more information about the dataset and its usage, see README_50_Hyperspectral_Reflectance_Images_2022.pdf and the published article https://dx.doi.org/10.1098/rspb.2021.2483.