Abstract
Human observers can recognize an object as small as thumb nail-sized photo or as large as several meter portrait photo. How do human observers realize this remarkable feat? What is the underlying neural mechanism? Some researchers suggested this is achieved by routing neural circuits processing different size objects to a size invariant neuron. However this seems logical circular since even though the researchers knew routing circuits processing different size objects, it is unclear how brain know this in the first place. Here we give a theory for size invariance. We argue that size invariance is logically equivalent to a unique representation for an object, and neural structures count more than learning at least for size invariance.
Human observers can recognize any different sized object after observing a specific size object. This is equivalent to a unique representation for an object. This can be proved mathematically easily. Then the problem becomes what is the unique representation? If we want to get a large obejct representation from a small object representation, some detail information (edges) is not available. On the other hand, we can get a small object representation from a large object representation. Therefore we suggest that the unique representation is the smallest representation for an object. To achieve this, any object projected on the retina is first processed by the nerual system to the smallest object the neural system can represents. Here the smallest object may be the smallest detectable by the human vision system.
Under this theory, all different size objects converge to this smallest object representation. This final convergent connection can be hard-wired by the brain due to neural structures representing the equivalence. This equivalent structure, however, may be partially built from evolution through observing object moving from different distances.
Chian Natural Sciences Foundation.