Digital Signal Processing Reference
In-Depth Information
Fig. 6.8
Augmented Game application [
27
]
Fig. 6.9
Panoramascope [
24
]
Blue, and Depth
) instead of just colour
(Red, Green, and Blue)
. The Gaussian
Mixture Model (GMM) components are assigned to pixels and learned from the
4-channel Image frame. The energy model is defined based on the foreground and
background histograms and the minimum energy represents good segmentation. The
segmentation is estimated using Graph Cut, which provides tentative classification
of pixels belonging to the respective classes. The above process is iterated until
convergence. The formulation of above mentioned process is briefly described be-
low. Let us consider image frame as an array
I
which includes both
R,G,B levels and depth values respectively. The segmentation is array of opacity
values
=(
I
1
,..
I
n
...
I
N
)
,
α
=(
α
1
, ...
α
N
)
at each pixel. 0 for background and 1 for foreground.
θ
is
