|M.Sc Student||Goldin Alexandra|
|Subject||A Model of Invariant Object Recognition in the Visual|
|Department||Department of Biomedical Engineering||Supervisor||ASSOCIATE PROF. Moshe Gur|
A striking feature of the visual system is its ability to represent visual information in a way that allows recognition to occur relatively independent of size, contrast, orientation, position on the retina, etc.
This ability has been studied for many years, and the emerging opinion is that object recognition is largely a hierarchical process such that as cells get further from the retina they code progressively more elaborate aspects of the visual stimulus.
The hierarchical hypothesis is attractive for its elegance and yet we believe that there are reasons to reexamine it. The implications of assuming hierarchical convergence leading to cells with large receptive fields representing single objects are contrary to perceptual, anatomical and physiological evidence. These implications have led us to suggest an alternative hypothesis for object representation and recognition.
We suggest that invariant object recognition is performed by a normalized representation, stored in memory as a prototype, which is a global transformation of the object representation in the primary visual cortex (V1). Normalization is performed by the many extrastriate brain areas that deal with tilt, magnification, and other global transformations.
The aim of this work is to construct a computational model of object representation and recognition by the visual system. The model will implement and test the feasibility of the normalized hypothesis.
The results indicate that the alternative hypothesis can explain recognition of objects in the visual cortex with out of hierarchical convergence and independent of the transformations that the objects were subjected to.