I am trying to generate 4 shapes that are equidistant in psychological similarity space - meaning that they are all equally discriminable from one another - which differ in 3 parameters, such that those parameters have a known mapping between objective and psychological space.

A simple example may help. We might define an object in terms of its width, height, and depth. The four objects A, B, C and D would have values as follows:

    height   width    depth

A.    2        2        1

B.    1        1        1

C.    2        1        2

D.    1        2        2

These points are equidistant in 3d euclidean space (they are actually the vertices of a tetrahedron). However, an increment of 1 in height may not be perceived psychologically as distinct as an increment of 1 in depth (in principle). More problematically, going from 1 to 1.5 may not be psychologically the same as going from 1.5 to 2 on any dimension, meaning that while these values are continuous, they are not linear in psychological space. In other words, objects equidistant in the metric of the parameters I'm using to generate them may not actually be equidistant in the eye/mind of an observer.

This is particularly true for this example, as point B is clearly more visually distinct from the others - it's the only cube!

I am therefore looking for 3 shape parameters where the mapping from objective space to psychological space is known, and in which there is not a loss of dimensionality in this mapping to below 3 dimensions in psychological similarity space.

I suspect this is not known, but I would like to avoid reinventing the wheel (not to mention running the complex pilot experiment this would require - it's not even my question of primary interest!)

  • 1
    $\begingroup$ Have you also considered rotation? That would make this much easier and also add a parameter that has well-defined limits (e.g., 0 degrees - 360 degrees). $\endgroup$
    – vizzero
    Jul 8, 2012 at 18:40
  • $\begingroup$ Thanks! Yes, I'll eventually be running a fairly complex design, with all the "normal" features varying. I think I've found a solution though - a good bit is known about the perceptual space of fourier descriptors - a way to characterize shapes. I've just now finished designing an experiment to determine the mapping from subjective to objective space with fourier descriptors. I will update people if this works! $\endgroup$
    – CHCH
    Jul 10, 2012 at 21:03

1 Answer 1


Consider the colour visual system. Take 3 monochromatic (in the physics sense) light sources of wavelength $420 \; nm$, $534 \; nm$, and $564 \; nm$; i.e. the peaks of spectral sensitivity for cones. Your 3 physical parameters are then 3 knobs $b$, $g$, $r$ that control the intensity of each light source as they shine on the same white surface. If you want more physical parameters, you can just add more knobs for other monochromatic lightsources, however 3 is sufficient. Colour perception is extremely well studied, so this comes as close as I can think of to a known mapping between between objective and psychological space. Unfortunately, you will run into some culture and language specific issues for judging the similarity of perceived colours.

  • $\begingroup$ I agree that color is probably the space I will need to consider. Thank you! But I am actually wondering about shape parameters in particular. $\endgroup$
    – CHCH
    Jul 5, 2012 at 19:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.