METU Informatics Institute
Seminar Series on
Cognition and Neuroscience
Neuroanatomy, fMRI and Modelling of
Human Vision
presented by
METU Informatics Institute
Cognitive Science Programme
March 10th, 2007, Saturday
Middle East Technical University
Cultural Convention Center, Hall D – KKM D
Although computational vision methods have been inspired from observations on actual human vision, as the image processing research developed, there has rather been less interaction between "Human Vision" and "Computer Vision". The aim of this seminar, is to help to bridge this gap by means of an introduction to human visual system followed by a series of talks which explore this interaction with various experimental and computational methods.
We hope that this will be of interest to participants both from engineering and cognitive backgrounds.
Participation is free, but to ensure adequate planning for venue and lunch, we do prefer prospective participants to pre-register online.
Thank you for your interest. Quotation for the Human Vision seminar registration for participants outside METU and outside IEEE has been filled. METU or IEEE members can still register by calling Ayse Ceylan at 2103743. Please leave your name, department and email address to her.
10.00: Opening remarks - D. Zeyrek
10.10-10.45: Human Visual System
D. Gökçay, Middle East Technical University
10.45-11.00: Questions and informal discussion
11.00-11.45: Retinotopic Mapping of Primary Visual Cortex
U.E. Emir, Boğaziçi Univeristy
11.45-12.00: Questions and informal discussion
12.00-13.00: Lunch Break
13.00-13.45: Mechanisms of Spatial Localization in the human visual system
Ö. Yılmaz, University of Houston
13.45-14.00: Questions and informal discussion
14.00-14.45: Cognitive Penetrability of Vision: Evidence from Spatial Frequency Processing
E. Özgen, Bilkent University
14.45-15.00: Questions and informal discussion
15.00-15.30: Coffee Break
15.30-16.15: Computational models of the primary visual cortex
A.B. Çürüklü, Sabancı University
16.15-16.30: Questions and informal discussion
16.30-17.15: Contextual and top-down effects in early vision
S. Tarı, Middle East Technical University
17.15-17.30: Questions and informal discussion
17.30: Concluding remarks - B. Say
Abstracts in the order of presentations
Human Visual system
Didem Gökçay
Informatics Institute, Middle East Technical University
In this talk, a general overview of the human visual system will be presented. The visual pathway is introduced starting from the rod and cone cells, continuing with responses of individual neurons in the lateral geniculate nucleus and primary visual cortex. At the lower levels of visual processing, edge-sensitivity of V1 cells and relevance of edge-sensitivity in natural scenes are discussed as well as operational principles of ocular dominance columns. At the higher levels of visual processing, interaction between vision and cognition is covered by explaining the 'what' and 'where' pathways and localization of 3-d representations in the brain. Cognitive modulation of vision is addressed briefly in the context of survival, by emphasizing attentional and limbic system effects on humans' perceptive abilities. Finally, visual systems of primates and humans are compared at the gross anatomical leve l, and disorders of the visual system are addressed very briefly. Overall, this talk aims to draw a general picture of the human visual system and provide a background for the forthcoming talks.
Retinotopic Mapping of Primary Visual Cortex
Uzay Emrah Emir
Institute of Biomedical Engineering, Boğaziçi University
Neuroimaging modalities such as EEG, PET, MEG and fMRI are the most common methods that neuroscientists use for functional brain imaging. By providing high resolution images, fMRI is used for functional brain mapping in clinic and research areas. Basically, fMRI is based on a response of the brain to a specific task which activates the specialized brain region. For instance, visual stimulus activates primary visual cortex. In animal studies, visual cortex was previously mapped by electrophysiological measurements. According to these studies, it is well understood that visual cortex can be mapped by polar coordinate system, retinotopy. In addition to this, results of retinotopy studies done by fMRI coincide with the electrophysiological measurements. Most famous method used in fMRI for this purpose is based on phase-encoded retinal stimulation and Fourier-based analyze of functional response of v isual cortex. With the help of this method, human visual areas V1, V2, VP, V3, and V4 can be precisely and non-invasively determined. In this study, how visual areas in primary visual cortex are determined by using fMRI will be briefly described.
MECHANISMS OF SPATIAL LOCALIZATION IN THE HUMAN VISUAL SYSTEM
Özgür Yılmaz
Electrical and Computer Engineering Department, University of Houston
Motion-Position Computation: Localization of objects is one of the fundamental functions of vision. The mechanisms for spatial localization of static and moving objects have been studied extensively in human vision. It has been suggested that the functional system for object localization is largely independent from the functional system responsible for the perception of objects’ motion. However, recent studies indicate interactions between the two systems (Whitney, 2002). We propose a model consisting of predictive projections from the motion system onto the position system. These predictive projections are suggested to increase the processing speed of the inherently slow position system in the expense of distortions in the visual space for flashed objects. Based on this model of motion-position interactions, we predict that the perceived position of a flashed visual stimulus should be attracted towards a nearby moving stimulus.
Attentive Tracking of Moving Objects: The world is a dynamic environment hence it is important for the visual system to be able to deploy attention on moving objects and to track their locations. In a visual task where it is instructed to track pre-assigned multiple moving objects (targets) among many distracter moving objects, observers can track up to five targets over several seconds (Pylyshyn and Storm, 1988). This paradigm is called Multiple Object Tracking (MOT). In an MOT task, attention is directed towards the target objects and the visual processing for these objects are enhanced; while it is shown psychophysically that the distracters are selectively inhibited to de-emphasize them from visual awareness (Pylyshyn, 2006). A two layer computational neural network model with integrate and fire neurons of synchronization capabilities is proposed and simulated for a basic attentive tracking task. Layer 1 in our model represents early visu
al areas that encode the position and basic features of objects, while Layer 2 represents the attention related brain regions that keep track of only the attended objects. In our model, inhibitory feedback from the second layer to the first layer causes de-synchronization with respect to a common oscillatory clock for distracter objects; and this prevents them to reach visual awareness and to interfere with attended target objects.
(Joint work with: Saumil S. Patel1,2,4, Srimant Tripathy1,3, Haluk Ogmen1,2: 1University of Houston, Electrical and Computer Engineering Department, 2 University of Houston, Center for Neuro-Engineering & Cognitive Science, 3 University of Bradford, Department of Optometry, 4 University of Houston, College of Optometry)
Cognitive Penetrability of Vision: Evidence from Spatial Frequency Processing
Emre Özgen,
Department of Psychology, Bilkent University
Early (low-level) visual processing has been thought of as immune to influences from higher level cognitive processes. Its contribution to visual analysis is characterised by a one way traffic where physiologically-determined mechanisms perform their function and send the information to higher levels of processing. Attentional mechanisms then filter the information as necessary and cognitive processes such as recognition, discrimination, and categorisation are performed.
Recent evidence however suggests that even at early stages, visual analysis might work in a two-way traffic system. Here, demands placed on the low-level mechanisms of visual processing by cognitive processes can influence or even determine the functioning of these mechanisms. In this talk, I will summarise evidence from our labs which support the latter interpretation, focusing on studies on one of the lowest levels of visual analysis, spatial frequency processing. Studies suggest that spatial frequency information in visual scenes and faces can be flexibly analysed depending on cognitive demands. In our labs we investigate whether such flexibility is indeed located in low-level visual processing stages. Our findings are consistent with an early locus for these effects.
COMPUTATIONAL MODELS OF THE PRIMARY VISUAL CORTEX
Ali Baran Çürüklü,
Computer Vision and Pattern Analysis Laboratory, Sabancı University
A canonical model of the primary visual cortex (V1) is presented. The main goal of this work is to develop mathematical models for understanding information processing in the brain. The V1 model is developed during exposure to (simulated) visual input. Learning rule that governed changes in the synaptic weights was the Bayesian Confidence Propagation Neural Network incremental learning rule. Connectivity pattern demonstrated by this correlation-based network model is similar to that of V1. In modeled cortical layers local horizontal connections are dense, whereas long-range horizontal connections are sparse. Layer 4 local horizontal connections are biased towards the iso-orientation domain, whereas long-range horizontal connections are equally distributed between all orientation domains. In contrast, both local and long-range horizontal connections of the layer 2/3 are biased towards the iso-ori
entation domains. The layer 2/3 network is axially specific as well. This V1 model demonstrates how the recurrent connections can be self-organized and generate a cortex like connectivity pattern. In both layers inhibition operates within a modeled hypercolumn. This is in line with what is found in the V1, i.e. inhibition is mainly local, whereas excitation extends far beyond the inhibitory network. Observe also that neither excitation nor inhibition dominates the network.
Based on this connectivity pattern the V1 model addresses several response properties of the neurons, such as orientation selectivity, contrast-invariance of orientation tuning, response saturation followed by normalization, cross-orientation inhibition. Configuration-specific facilitation phenomena are explained by the axially specific layer 2/3 long-range horizontal connections. It is hypothesized that spike and burst synchronization might aid this process.
CONTEXTUAL AND TOP-DOWN EFFECTS IN EARLY VISION
Sibel Tari
Dept. of Computer Engineering, METU
I will propose a coupled set of Partial Differential Equations to model contextual and top-down influences during de-noising and edge enhancement. Local computations separated in time or space interact in a way that the large scale effects compete with small scale effects to yield perceptually preferable filtering results. Experimental results verify significant improvement in filtering under difficult noise cases, nonuniform contrast and existence of textured areas.
Time permitting, I will talk about how non-metric nature of human similarity judgements can be modeled by permitting contextual influence.