Naslov
A neural model of contour tracing operator

Autori
Domijan, Dražen ; Marić, Mateja

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Skup
43rd European Conference on Visual Perception ECVP 2021

Mjesto i datum
Online, 22.08.2021. - 27.08.2021

Vrsta sudjelovanja
Poster

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
contour tracing ; object-based attention ; perceptual grouping ; winner-takes-all

Sažetak
Contour tracing is a visual routine that enable serial grouping of distant contour segments into a unified object representation, and segregation of selected object from distractors. At the neural level, contour tracing is implemented by spreading of increased firing rates among orientation-tuned units in the primary visual cortex (Pooresmaeili & Roelfsema, 2014, Current Biology, 24, 2869-2877). Here, we developed a neural network model for contour tracing whose dynamics is comparable to neurophysiological measurements. Activity spreading is achieved by computing activity difference between two sets of units: 1) attentional units whose horizontal connections enable propagation of activity enhancement to collinear neighbours and 2) non-attentional units that receive only feed-forward input. The model employs multiple spatial scales in order to account for variable speed of contour tracing. In addition, the L-junction detection network enables tracing to change direction to non-colinear orientations when L-junction is reached. Computer simulations showed that the model correctly predicts slower tracing times when target and distractor contour were placed closer to each other and when they intersected at smaller angle as it was observed experimentally. In addition, the model is capable of tracing contours whose segments vary in contrast.

Izvorni jezik
Engleski

Znanstvena područja
Psihologija, Kognitivna znanost (prirodne, tehničke, biomedicina i zdravstvo, društvene i humanističke znanosti)

POVEZANOST RADA