Psychometrically scaled image sets

The dataset consists of Three noisy measurements ("leaf2.png", "down2.png", "terrain2.png") used to construct the sample sets ("leaf2"/"down2"/"terrain2" -"_sample_set.zip"). The sets contain total variationally denoised solutions of the noisy measurement for a grid of regularization parameter values. The grid is discretized by the solutions' HaarPSI values so that any two consecutive images from each set will have a similarity value of at least 0.990. Two psychometrically scaled sets ("leaf2_psychometric_set.zip", "terrain2_psychometric_set.zip"). These are subsets of the sample sets, discretized by the solutions' HaarPSI values so that the threshold for the former set is 0.984, and for the latter 0.956. The sample set "down2_sample_set.zip" can be used as is in lieu of a psychometrically scaled sets, keeping in mind the possibility that the threshold value is higher than 0.990. A list of the parameters used to denoise each solution. The algorithm used for denoising the measurements. The code is written by Emilia Blåsten and Lílian Ferreira de Freitas and it is based on the Chambolle-Pock algorithm commonly used for total variation denoising and regularization. The psychometrically scaled datasets are optimized for visual testing in mathematical and imaging research. Each set contains as few perceptually similar images as possible, preventing the collection of redundant or skewed data, and significantly reducing the time and resources required for comparison testing. Note: The sample sets can be scaled in multiple ways. The one presented here is only one possibility of doing so.