08.02.2013 Views

Bernal S D_2010.pdf - University of Plymouth

Bernal S D_2010.pdf - University of Plymouth

Bernal S D_2010.pdf - University of Plymouth

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

2.3. ILLUSORY AND OCCLUDED CONTOURS<br />

established directions, and is considered one <strong>of</strong> the milestones <strong>of</strong> perceptual organization. A<br />

similar concept, known as relatability, has been suggested as the guiding principle for contour<br />

completion in the visual system. It is based on smtwlhness principles such as contours being<br />

differentiable at least once, monotonic, and having bending angles below 90" (Kellman 2003).<br />

An important tool to explain contour completijin is the bipole (Neumann and Mingolla 2001.<br />

(irossberg et al. 1997. Roelfsema 2006). Although it can have diverging definitions depending<br />

on the author, most generally its function is to evaluate the effect a contour element at a given<br />

location has on the likelihood <strong>of</strong> perceiving a contour at a second k)cation. The bipole is typ~<br />

ically represented by a characteristic hgure-eighl shape which describes the coupling strength<br />

between the centre unit and surrounding units according to their relative position and orienta­<br />

tion. A neural implementation <strong>of</strong> this concept gives rise to bipole cells, defined as nonlinear<br />

grouping operators which receive input from real edges falling inside the bipole lobes (see Fig­<br />

ure 2.12). It is strongly grounded on anatomical, psychophysical and physiological data (see<br />

Neumann and Mingolla (2001) Cor a review), and provides a biologically grounded method <strong>of</strong><br />

implementing good continuation and relatabilily principles.<br />

This section discusses existing approaches and models that show how these geometric concepts<br />

can be implemented with neural mechanisms and how they relate to the physiology and anatomy<br />

<strong>of</strong> the visual cortex.<br />

2.3.2.3 Clas.sjficati(in or theoretical models<br />

Two broad theoretical categories for contour completion models have been considered. The first<br />

one relies on feedforward processing, and is also known as base grouping. The second deals<br />

with recursive mcxlels, placing the focus on lateral and feedback connections, and is also known<br />

as incremental grouping (Roclfsema 2006, Neumann and Mingolla 2001),<br />

An alternative, though compatible, classification focuses on the specific mechanisms involved<br />

ill contour completion, and proposes three broad classes: 1) contour interpolation, which in­<br />

wardly extends two aligned line segments; 2) contour extrapolation, which outwardly extends<br />

a segment <strong>of</strong> a single line segment; and 3) figural feedback, whereby a high-level representa­<br />

tion feeds back to complete missing contours (Halko et al. 2{K)8). The lirst two classes can<br />

54

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!