Information Technology Reference
In-Depth Information
independent systems were proposed, using the principal component analysis that
allows for uncorrelated components. The two commonly used systems are:
•
The system I
1
I
2
I
3
of Ohta
et al.
[28] defined by
I
1
=
R
+
G
+
B
3
,
−
I
2
=
R
B
,
(19)
2
I
3
=
2
G
−
R
−
B
.
4
•
The system H
1
H
2
H
3
whose components are given by
H
1
=
R
+
G
,
H
2
=
R
−
G
,
(20)
R
+
G
2
H
3
=
B
−
.
Both color spaces are a linear transformation of the RGB system.
3.2
A Survey of Color Based Stereo Methods
It is known that, in the human visual system, binocular vision is a key element of
the three-dimensional perception: each eye is a sensor that provides to the brain its
own image of the scene. Then, the spatial difference between the two retinal images
is used to recover the three-dimensional (3-D) aspects of a scene. The experiment
conducted in [29] on nine subjects demonstrate that the amount of perceived depth
in 3-D stimuli was influenced by color, indicating therefore that color is one of
the primitives used by the visual system to achieve binocular matching. This study
confirms that color information may be used to solve the stereo matching problem.
Different techniques proposed until now to deal with color stereo matching will be
presented below. We will first describe color-based local methods and then we will
focus our attention on global methods.
3.2.1
Local Approaches
We exposed in Section 2.3.1 the principle of local approaches to solve the corre-
spondence problem based on gray level stereo images. The main idea is to perform
a similarity check between two equal sized windows in the left and right images. A
similarity measure between the pixel values inside the respective windows is com-
puted for each disparity
d
within a search range
, and the disparity providing the
minimum value is regarded as the optimal disparity value. The correlation measure
is defined for gray value images as follows:
Ω
