Guillermo Aguilar, BCCN Berlin / GRK 1589 / TU Berlin

On the use of MLDS in the study of depth and lightness perception

An open question in vision research is how to measure the perceptual dimension evoked by the stimulus in a reliable way. Although a variety of psychophysical procedures are available, it is still a challenge to find methods that are efficient and avoid critical confounds, such as strategies triggered by difficult and unnatural tasks used by discrimination methods. In this doctoral thesis I propose the use of Maximum Likelihood Difference Scaling (MLDS, Maloney & Yang, 2003) as a reliable tool for measuring perception. MLDS is a method based on judgements of appearance of clearly visible stimulus differences in an easy and intuitive task, and it allows the estimation of perceptual scales in an efficient way.

Here I first use numerical simulations to test the accuracy and precision of the scales derived with MLDS, and I also tested the effect of violations of the model assumptions. The results of these simulations establish the validity of MLDS as a method for measuring appearance. Then, we evaluated MLDS experimentally in the domain of lightness perception. We measured perceptual lightness scales under different viewing conditions and we validate the derived scales empirically by predicting lightness matches that were derived in a classical asymmetric matching task. A large practical benefit of MLDS is that it renders the task easy for the subject and thus minimizing the potential influence of strategies. At the same time the perceptual scales provide a more direct estimate of internal variables against which theoretical models of appearance can be tested.

In a third part I study the relationship between MLDS and discrimination methods as suggested by Devinck & Knoblauch (2012). In simulations MLDS was more efficient than the traditional 2-AFC discrimination method while at the same time providing analogous sensitivity estimates. I also tested this equivalence experimentally in a slant-from-texture task, for which sensitivity has been previously studied in the literature. Here I found varying degrees of equivalence and it remains to be tested in the future whether these differences are due to true differences in the perceptual representation, or to violations of the model assumptions.

Together with the use of realistic stimuli, MLDS offers a reliable method to measure the perceptual dimension, and in that way enabling the testing of theoretical models of perceptual inference.

Additional Information

PhD defence as part of the GRK 1589 "Sensory Computation in Neural Systems"

Organized by

Marianne Maertens / Robert Martin