I am conducting a study that will present participants with categories of stimuli to determine the influence of stimulus valence on attentional processes. To avoid low level confounds, it is important that the stimuli be matched on physical attributes, such as luminance, contrast, and spatial frequency. Any advice for yielding these indices would be very much appreciated.
If you can characterize / calibrate your display, it should be possible to convert the value of each pixel to a luminance level. There are different ways to calculate contrast, but what seems to affect behavior most consistently is simply the standard deviation over the luminanve values (note that Michelson or Weber contrast don't really work for natural stimuli), so that could be a first contrast measure to control. There is a paper by Peli who proposes a more complex measure, which might be worthwhile to check out. Color contrast may also affect gaze, and to control for this, you would want to convert (RGB) values to some color space where color is independent of luminance. We usually use DKL space which is also supposed to be physiologically correct. (Luminance and luminance contrast can also be corrected for in DKL space independent of any color corrections.) I've never done any corrections on spatial frequency, but a 2D FFT performed on a luminance-only version of your stimuli should be a good start. HTH.
BTW, similar study from our lab: Naber, Hilger & Einhäuser (2012): www.journalofvision.org/cgi/pmidlookup?view=long&pmid=22281692
By valence do you actually mean categorisation of the image content such as 'positiveness' or do you mean how much the image 'stands out', that is, conspicuity?
Not sure if it helps, but I am doing a set of experiments on (basically) pop-out while attempting to keep the stimuli 'natural'. We use 1/f noise which we then (Gaussian) filter in the Fourier domain, and finally calculate the frequency band of the final product for a given monitor/viewing distance.
I assume that if you are matching luminance, contrast and spatial frequency, that the images are not actual real-world images as that would be very difficult as even with very broadband SF filtering, filtered natural images are perceptually quite different from the originals and at particular bands may render very different images to be quite similar (for example two faces, with all the high SF content they are quite different whereas with only the low SF information they may look very similar).
Have you looked into Gabor stimuli? They are great for really controlled experiments as they have a very specific luminance, contrast, SF profile.
Can't wait to see what you come up with!
Alex
Thank-you for your responses.
To clarify, the program of research will utilise a range of pictorial stimuli. Earlier studies will use complex scenes, which we are attempting to match on contrast and brightness. Later studies will likely include photographs of faces. The stimuli will vary systematically in terms of their affective valence – threatening, pleasant, and neutral.
I have stumbled upon the SHINE toolbox for MATLAB (Willenbockel et al., 2010; doi: 10.3758/BRM.42.3.671), which may assist in matching the stimuli on key image properties. However, I am still looking for a program that will simply provide indices of brightness and contrast for each image. At present I am able to yield an index of luminosity using the histogram window in Adobe Photoshop – however I am doubtful whether this is an appropriate measure to use as an index of brightness.
I suppose you could convert your images to grayscale and report the average luminance? Though with natural images that statistic is probably useless. Perhaps you could fit the histogram of luminance values to a particular distribution, then use the parameters of that distribution to make future comparisons?
- Badges
- Science topic