A toolbox and sample object perception data for equalization of natural images

Abstract

For psychologists and neuroscientists, careful selection of their stimuli is essential, so that low-level visual features such as color or spatial frequency do not serve as confounds between conditions of interest. Here, we detail the Natural Image Statistical Toolbox, which allows scientists to measure, visualize, and control stimulus sets along a set of low-level visual properties. Additionally, we provide a set of object images varying along several perceptual object properties, including physical size and interaction envelope size (i.e., the space around an object transversed during an interaction), serving as a test-bed for the Natural Image Statistical Toolbox. This stimulus set is also a highly characterized set useful to psychology and neuroscience studies on object perception.

Keywords: Natural image statistics; Object interaction envelope; Object perception; Spatial frequency.

Fig. 1

An example of how one might use the Natural Image Statistical Toolbox to compare stimulus conditions (in this case, the conditions from the Object Interaction Envelope Stimuli set). (Left) Visualizations of spectral energy levels of the different conditions orthogonalizing object physical size and envelope size . As one can see, there are subtle differences in the spatial frequency information across the conditions. The Natural Image Statistical Toolbox also outputs statistical information and performs t-tests between relevant conditions. In this case, there were no significant differences in spatial frequency for comparisons of interest (i.e., between orthogonalized conditions, and between intercorrelated conditions). (Right) The image average of all of the stimuli for each condition. With this, one can assess condition differences in shape, spatial frequency information, color, orientation, amount of white space, etc.

Fig. 2

The 360 object images included in the Object Interaction Envelope Stimuli set. Object properties for all small-scale object images are also included.