Computational Imaging Group

TAMPERE21 noise-free image database

1000 color images 1200x800, 960x640 and 768x512 pixels without noise (variance < 1), without interpolation and without lossy compression.

Images and image details (see tampere21.mat): tampere21.zip (1.6 Gb).

Please refer to the following paper:
S.G. Bahnemiri, M. Ponomarenko, K. Egiazarian, "Learning-based noise component map estimation for image denoising", IEEE Signal Processing Letters, 2022, 5p.

Short description:

Images for the database are obtained with monitoring of noise level (the photo shooting was carried out on calibrated camera Canon EOS 250D).

Depending on ISO value, central part of each image was cropped and downscaled in 5, 6 or 7 times without interpolation (by averaging of pixel values for regions 5x5 pixels, 6x6 pixels or 7x7 pixels respectively). Such averaging provides decreasing of noise variance on images in 25, 36 or 49 times respectively and grants that the variance of noise in the images of Tampere21 is smaller than 1.

For all images of Tampere21 the following parameters of source images were stored in the database (see tampere21.mat): ISO level, camera name, focal length, exposure time, f-number, white balance mode, downscaling factor (5, 6 or 7).