See the gallery of stimuli.
This data set was measured for comprehensive testing of visual models using actual distortions and complex images. The distortions are white-noise, JPEG compression artifacts, bilinear upsampling (blurring), and sine-gratings. All distortions are restricted to a small image regions to simplify the measurements.
The thresholds were measured in the 4-alternative-forced-choice experiment, in which all four images were shown side-by-side. The measurement procedure was the same as for the "Contrast sensitivity for wide luminance range" data set.
This is the original Visual Difference Predictor based on corresponence with the author and the book chaper: Daly, S. "The Visible Differences Predictor: An Algorithm for the Assessment of Image Fidelity." In Digital Images and Human Vision, edited by Andrew B. Watson, 179-206. MIT Press, 1993.
The metric uses the default parameters from the paper except the masking slope, set to 0.9 for all bands, which was found from fitting the metric to the masking data sets. One optional component in the VDP is computing contrast in the cortex filtered images, which could be either global or local. This version uses global contrast.
The metric also includes an improved variation of the phase uncertainty, as described in: Lukin, A. "Improved Visible Differences Predictor Using a Complex Cortex Transform." In International Conference on Computer Graphics and Vision (GraphiCon), 2009. Phase uncertainty is not mentioned in the '93 book chapter, but is described in the patent application. The method proposed by Lukin achieves the same goals as phase uncertainty but is more elegant and efficient than the approach described in the patent.