Image Processing Reference
In-Depth Information
Chapter 6
Visual Attention Modelling in a 3D Context
Haroon Qureshi and Nicolas Tizon
Abstract This chapter provides a general framework for visual attention
modelling. A combination of different state-of-the-art approaches in the field of
saliency detection is described. This is done by extending various spatial domain
approaches to the temporal domain. Proposed saliency detection methods (with and
without using depth information) are applied on the video to detect salient regions.
Finally, experimental results are shown in order to validate the saliency map quality
with the eye tracking system results. This chapter also deals with the integration of
visual attention models in video compression algorithms. Jointly with the eye
tracking data, this use case provides a validation framework to assess the relevance
of saliency extraction methods. By using our proposed saliency models for video
coding purposes, we demonstrate substantial performances gains.
6.1
Introduction
With the advancement in modern digital technologies, the need for better quality
assessment of 3D multimedia contents has also increased. While delivering 3D
media contents to fixed and mobile users, bandwidth, equipment performance, and
cost are dominant factors. These requirements can be minimized with the help of
newly emerging technologies in the field of compression, intelligent video coding,
and most specifically in the field of visual attention modelling. For example, one
possible way is to compress the video equally without considering the contents
which requires a specific amount of bandwidth. Another but a smarter way is to use
an encoder that is sensitized to encode salient regions with higher data rate. This can
be done by exploring visual attention modelling techniques.
H. Qureshi (
*
)
Institut f¨r Rundfunktechnik GmbH (IRT), 80939 Munich, Germany
e-mail:
qureshi@irt.de
N. Tizon
VITEC Multimedia, 99 rue Pierre S´mard, 92324 Chatillon, France
e-mail:
nicolas.tizon@vitecmm.com