Image Processing Reference
In-Depth Information
Fig. 4.8 View selective
blending extraction
Fig. 4.9 Patch size in view
extraction
(by increasing its distance from the center) before the blending. The value of this
drift controls the depth plane on which the resulting extracted view will be
focused.
4.2.4.3 Single-Sized Patch
This method proposed by Georgiev is based on the characteristics of the focused
plenoptic camera [
8
](seeSect.
4.2.1.2
) for which there are both angular and
spatial information within one MI. It uses the same steps as the basic extraction
illustrated in Fig.
4.7
, but instead of extracting one pixel per MI, a patch
(corresponding to a square zone of pixels) is extracted. The angle of view chosen
depends on the relative position of the center of patches in the MIs. The size of the
patch defines the depth plane in the scene on which the extracted view will be
focused: the larger the patch is, the closer the focus plane is. The objects that are
further or closer will present the following artifacts, illustrated in Fig.
4.9
.Ifthe
patch is too large (i.e., the object is too far), then redundant parts of the object will
be represented in several adjacent patches. If the patch is not large enough (i.e.,
the object
is too close),
then parts of the object will not be represented
(pixelation).
4.2.4.4 Single-Sized Patch Blending
As for the previous method, one patch per MI is extracted. However, the pixels
outside the patch
s borders are kept to be blended. The blending is done here with a
weighted averaging. The weight of pixels inside the MI can be attributed using a
Gaussian distribution centered on the patch, so that the pixels that are further from
the center have a lighter weight as illustrated in Fig.
4.10
. The blending allows
'
