Graphics Reference
In-Depth Information
N
=
ˆ
R
=
W
I
(8)
ij
ij
,
k
ij
,
k
k
1
k
th
I
where
the
input image in the sequence.
k
However, applying equation (8) directly will produce an unsatisfactory result.
Wherever weights change quickly, disturbing seams will appear. To address the seam
problem, this paper uses a technique inspired by Burt and Adelson [16]. Their original
technique seamlessly blends input images guided by an alpha mask, and works at
multi-resolutions using pyramidal image decomposition. This technique is adapted to
our case, where there are
N
N
images and
normalized weight maps that act as alpha
th
A
l
mask. Let the
level in a Laplacian pyramid decomposition of an image
be
l
l
G
B }
B
L
A}
defined as
, and
for a Gaussian pyramid of image
. Then, the coeffi-
cients are fused in a similar fashion to equation (8):
N
=
ˆ
l
ij
l
l
L
R
=
G
W
L
I
{
}
{
}
{
}
(9)
ij
,
k
ij
,
k
k
1
l
of the Laplacian pyramid is computed as a weighted
average of the original Laplacian decomposition for level
For example, each level
th
l
level of
Gaussian pyramid of the weight map serving as the weights. Finally, the pyramid
l
, with the
l
L
R }
R
. An overview of this framework is given in figure 2.
For dealing with color images, each color channel will be fused separately.
is collapsed to obtain
Fig. 2. Pyramid collapsing
5
Experiments Analysis and Evaluation
In order to test the proposed multi-focus capture and fusion system, several data sets
have been captured by the device. Each data set has different focuses and its resolu-
tion is 2144×1424 pixels. The proposed fusion method is compared with other multi-
focus fusion methods such as average method and wavelet method [17-19]. In the
experiments, standard deviation, information entropy and average gradient are used to
 
Search WWH ::




Custom Search