Coding the Presence of Visual Objects in a Recurrent Neural Network of Visual Cortex

Bevor wir in der Lage sind Sehobjekte zu erkennen, müssen wir diese von ihrem Hintergrund trennen. Dies bedarf eines schnellen Mechanismus, der feststellt ob und an welchem Ort ein Objekt vorliegt - unabhängig davon um was für ein Objekt es sich handelt. Vor wenigen Jahren wurden Kantenzugehör...

Full description

Saved in:
Bibliographic Details
Main Author: Zwickel, Timm
Contributors: Eckhorn, Reinhard (Prof.Dr.) (Thesis advisor)
Format: Doctoral Thesis
Published: Philipps-Universität Marburg 2006
Online Access:PDF Full Text
Tags: Add Tag
No Tags, Be the first to tag this record!

Before we can recognize a visual object our visual system has to segregate it from its background. This requires a fast mechanism for establishing the presence and location of objects independent of their identity. Recently, border-ownership neurons were recorded in monkey visual cortex which might be involved in this task Zhou et al. (2000). Border-ownership neurons respond with increased rates when an object surface extends to one specific side of the contour they encode. Conversely, the rate decreases when the contour belongs to an object extending to the other side. This selectivity for object position relative to a contour is called border-ownership. Zhou et al. (2000) found border-ownership neurons that encode oriented contrast edges or lines in areas V1, V2, and V4 of visual cortex in awake monkeys. In order to explain the basic mechanisms required for fast coding of object presence I developed a neural network model of visual cortex consisting of these three areas: - Area 1: encoding orientation contours - Area 2: encoding curvatures - detecting the presence of stimulus objects In my model feed-forward and lateral connections support coding of Gestalt properties including similarity, good continuation and convexity. Model neurons of the highest area (Area-3) respond to the presence of an object and encode its position, invariant of its form. Feedback connections from Area-3 to Area-1 facilitate orientation detectors activated by contours belonging to potential objects, and thus generate the experimentally observed border-ownership property. Border-ownership feedback is transmitted directly to neurons encoding convex contours of an object and indirectly via lateral connections into concavities. My simulations show that the border-ownership connections of my model can be learned with Hebbian learning. This confirms my networks architecture. In conclusion, my network is an encompassing model bringing together several aspects of object detection and coding. The model reproduces the experimental observations of border-ownership by Zhou et al. (2000). Further, border-ownership feedback control acts fast and significantly improves the figure-ground segregation required for the consecutive task of object recognition.