PSYC 212: Object Recognition I

Difficulties in Object Recognition Systems

Context, lighting conditions, and variability of objects

Template Theory

A theory of object recognition the visual system recognizes objects by matching the neural representation of the image with a stored representation of the same object (Works with poor variability)

Exemplar Theory

A theory of object recognition that we compare the objects we see to a variety of 'exemplars' that make up groups in our subconscious (Works with high variability)

Generated Context Model (GCM)

A mathematical proof of the exemplar theory proposed by Nosovsky

Prototype Theory

A theory of object recognition where we compare what we see to an abstract, 'ideal' version of what we might be looking at (Works well with medium variability)

General Recognition Theory (GRT)

A theory of object recognition where categorization is based on multivariate signal detection focusing on how perceptual distributions influence decision-making (Categories based on statistical boundaries, Ashby)

GRT vs. GCM

GCM based on similarities, GRT based on statistically defined boundaries concerning physical dimensions

Recognition by Components Theory

A specific view of an object can be represented as an arrangement of simple, universal 3-D shapes called geons (Bieterman)

Grandmother Cell Theory

Theory that there is a particular cell in the ventral processing stream whose job is to fire when you see a particular object or person (such as your grandmother) (Jerry)

Deep Neural Network (DNN)

A type of "machine learning" in artificial intelligence in which a computer is programmed to learn something (here object recognition). First the network is "trained" using input for which the answer is known ("that is a cow"). Subsequently, the network can provide answers from input that it has never seen before.

Early Level Vision

Retinal Ganglion Cells, LGN, Optic Tract, V1

Intermediate Level Vision

V2, V3, V4, etc.

High Level Vision

IT Cortex, FFA, PPA, etc.

Object Recognition Hierarchy

One process leads to another, and signals are sent from one area to another that has more strength

Intermediate Level Vision Purpose

Grouping object features by contour, textures, and surfaces

High Level Vision Purpose

Recognizing complex shapes and categories

Computerized Edge Detector Effectiveness

Computers are worse at detecting edges than the human eye because of contrast ability and differences in contextualization

Illusory Contour

a contour that is perceived even though nothing changes from one side of it to the other in an image (e.g undefined triangles in between other shapes)

Gestalt Theory

The whole is greater than the sum of it's parts

Gestalt Principles for Object Identification

Proximity, similarity, good continuation, closure, common fate, common reign, figure-ground, and connectedness

Common fate (Gestalt)

Humans tend to perceive elements moving in the same direction as being more related than elements that are stationary or that move in different directions.

figure-ground

the organization of the visual field into objects (the figures) that stand out from their surroundings (the ground).

Five Principles of Intermediate Vision

Grouping, Seperation, Prior Knowledge, Minimize Ambiguity, Avoid accidents (illusions)