A computational study of visual template identification in the SAIM: a free energy approach

Yahya, Keyvan (2013). A computational study of visual template identification in the SAIM: a free energy approach. University of Birmingham. M.Phil.

Preview

Yahya13MPhil.pdf
PDF - Accepted Version
Download (3MB)

Abstract

This thesis aims to understand how humans could recognize and identify objects. Our main method for doing so is developing a computational model of recognition/ identification process. This work will account for the process of visual object identification which usually takes place in multiple environments including various objects. Since we assumed that visual selective attention is central in disambiguating of objects, the results of our work will include an implementation of what visual selective attention does. Initially this thesis will draw on two successful approaches to human information processing. On one hand, we will base our work on the Selective Attention for Identification model(SAIM). The SAIM combines visual selective attention and object recognition. On the other hand, I will use the "Free Energy" approach proposed by Karl Friston to implement the fundaments of SAIM and expand it by incorporating an identification process. We will then reason for our cliam that holds that perceptual recognition, attention and identification minimizes the "surprise" (prediction error) about incoming sensory signals (Friston, 2006). It will be demonstrated that identification process would lead to an unsupervised extraction of object templates (prior beliefs about the causes of sensory input) from a series of multiple visual scenes to execute a successful object recognition task. At the end of our work, we would test our model by doing a series of computational experiments which are performed in Matlab environment consisted of various neural networks.

Type of Work:

Thesis (Masters by Research > M.Phil.)

Award Type:

Masters by Research > M.Phil.

Supervisor(s):