Multi-scale perception model for visual illusion on hybrid image

In this paper, we propose a neural model of multi-scale perception with so-called hybrid images that two different interpretations of a picture are perceived by changing the viewing distance. In this model, 2D-Gabor-based wavelets are prepared as features observed in visual cortex or retina, and for feature similarity computations by finding multi-scale correspondences between the input hybrid image and its model picture. We show that the feature similarities can also vary by changing the input image resolution (corresponds to the viewing distance). In discussion, we indicate that our model proposed here is one of the neural models that potentially support human visual recognition process.