Image Processing Reference
In-Depth Information
Visual attention modelling is a method that tries to estimate features in a visual
scene that will catch the viewer
s attention. The competency of human visual
system to notice visual saliency is enormously fast and dependable. Visual saliency
is represented by a saliency map. Saliency map can be defined as a visual repre-
sentation of a corresponding scene.
Computational models of visual attention use different image features such as
color, intensity, orientation, face, etc. Those are used as a clue to firstly identify
salient regions and then predict places that are likely to attract human attention. The
following section presents an overview of different saliency detection techniques.
'
6.2 Overview of Saliency Detection Approaches
The capability of computational image processing has increased up to an extent that
it has opened many new doors of research possibilities. Since many years, detecting
salient regions is an important research topic in the area of visual attention model-
ling. It provides a fast initial pre-processing stage for many vision applications. The
recent developments in the field of visual attention modelling are greatly inspired
by two famous models regarding visual perception that try to mimic a human visual
system model. One of the popular models is Feature Integration Theory (FIT)
[
35
]. An extension to FIT model is Guided Search (GS) [
36
]. Both models attempt
to explain schematically a human
s behavior while looking at a visual scene. The
idea of automatic saliency detection is also equivalent to these models. A very
comprehensive survey, covering an analysis of scores, datasets, and a model of
state-of-the-art technologies in visual attention modelling is presented in the papers
[
3
,
6
].
Generally saliency detection methods can be categorized into two parts: image
based saliency and video based saliency approaches. Image based saliency methods
emphasize to find salient regions from the background. While the video based
saliency methods aim is to distinct salient motions from the background. Detection
of salient regions can be done in many different ways:
'
• Detection using various clues
• Detection using spatial information
• Detection using temporal information
• Detection using depth information
6.2.1 Detection Using Various Clues
The ideas of saliency detection are commonly used in many applications such as
object-of-interest segmentation [
16
,
24
], motion detection, frame rate
up-conversion [
21
,
22
], image re-targeting [
31
], person tracking [
10
], identification
