Image Processing Reference
In-Depth Information
[
13
], video summarization [
27
], compression [
15
], and also in the area of energy
saving [
4
].
With the development of new applications, researchers are trying hard to make
the saliency detection techniques as much robust as possible. That is why in order to
detect salient region, different clues for example, camera motion [
1
], face [
7
,
8
,
29
],
and speech are used by the researchers.
6.2.2 Detection Using Spatial Information
Detecting region of interest (ROI) from the intra image information has been
studied since many years. Many different approaches have proposed in the mean-
while. Koch et al. [
25
] proposed to extract maps for each feature and apply Winner
Take All (WTA) procedure in order to highlight the most apparent area of interest.
It considers bottom-up criteria due to their low level complexity. They include three
features based on the difference in terms of color, intensity, and orientation. These
features are the most attractive elements for the human brain. Figure
6.1
presents
three images with a specific point of interest based on these features and Fig.
6.2
illustrates this approach in more detail.
Many approaches for saliency detection emerge from the field of communica-
tions engineering. Itti et al. simulated the process of human visual search in order to
detect salient regions [
26
]. Achanta et al. and Cheng et al. estimate saliency using
Fig. 6.1 Most attractive features for human brain. From
left
to
right
: color, orientation, and
intensity
Fig. 6.2 Overview of
Winner Take All (WTA)
Approach
