10/19/2012 4:30:00 PM
Chapter 3 Cog Lecture
Recognizing an object means doing two things.
First, you have to generate some internal representation of the object - a description of
the object in the language of the nervous system
Second, you have to locate a matching object, with the same description, in memory
Why is object recognition difficult? In terms of our retinal images
1) Orientation (objects that are same may have diff retinal images)
2) Ambiguity (crumpled toilet paper and cream cheese)
3) Incompletion (partially blocked, bad lighting, etc)
4) Ambiguity of the retinal images (diff shapes but same retinal
Example of picking up a cup and turning it around
you don't perceive the object as changing shape, only as changing orientation
Suppose you found yourself in a very unfamiliar situation
Your analysis of the object would be “bottom-up” – focusing on its shape, color, surface
texture, weight, and other such sensory qualities.. JAMES GIBSON believed in bottom up
processing as a way to perceive and recognize objects (not a dominantly shared view)
In a bottom-up model, the internal representation of the stimulus is the critical thing.
With template models, how would you learn to recognize new objects
Feature models – David Hubel 1960 and Torsten Wiesel .. used a single-cell recording technique to study
the visual system in cats… horizontal lines got reactions while sometimes just vertical lines did. NOBEL
They found that different neurons responded to different basic features.
A Problem - how can the cup be represented in memory as a particular list of features if
the features depend upon its orientation with respect to the viewer?
A second problem is that any visual object consists not just of a list of features but also of a
plan of sorts for how the features are arranged spatially with respect to each other
A final problem with feature