Reproducing the Muller-Lyer illusion in a computational feed-forward object recognition model

The Muller-Lyer illusion is where the perceived length of lines is altered when the shafts are terminated by various fins. Lines appear shorter in a 'wings-in' configuration versus longer in a 'wings-out' configuration. Explanations for this effect range from low level signal processing to the misapplication of rules from higher cognitive areas. Computer models that mimic visual processing allow for some of these proposed contributing factors to be tested in isolation. The HMAX model is a current state-of-the-art object recognition model that is also biologically plausible [Mutch and Lowe, 2008 International Journal of Computer Vision 80(1) 45-57]. We trained this model to perform a dual categorisation task based on relative line lengths within an image. We then measured the accuracy of the system in categorising control images versus illusory images. Our results indicate this feed-forward model replicated an overall illusory effect.1 page(s