Global error minimization in image mosaicking using graph laplacians and its applications in microscopy
An image mosaicking method includes performing pairwise registration of a plurality of tiles ( 101 ), determining absolute homographies for each of the plurality of tiles according to the pairwise registration ( 102 B), and performing a blending of the plurality of tiles to obtain a stitched image according to the absolute homographies ( 103 ).
1. A non-transitory computer-readable storage medium with an executable program stored thereon, wherein the program instructs a microprocessor to perform an image mosaicking method, the method comprising:
performing pairwise registration of a plurality of tiles;
receiving prior homography information for each of the plurality of tiles;
determining absolute homographies for each of the plurality of tiles according to the pairwise registration by minimizing a following objective function
E
(
p
1
,
…
,
p
N
)
=
1
2
∑
i
=
1
N
∑
j
∈
Nbd
(
i
)
p
i
-
p
j
-
p
~
ij
2
+
λ
2
∑
i
=
1
N
p
i
-
p
~
i
2
,
wherein N is a number of tiles, Nbd(i) denote a set of tiles that share any overlap with tile i, p i denotes a homography of tile i, {tilde over (p)} ij denotes a translation obtained by applying the pairwise registration between neighboring tiles i and j, {tilde over (p)} i denotes the prior homography of tile i, and λ denotes a predetermined weighting for the prior homography; and
performing a blending of the plurality of tiles to obtain a stitched image according to the absolute homographies.
2. The computer-readable storage medium of claim 1 , wherein performing the pairwise registration comprises maximizing a similarity measure.
3. The computer-readable storage medium of claim 1 , wherein determining the absolute homographies for each of the plurality of tiles comprises revising the pairwise registration according to an objective function for an edge incidence matrix of the plurality of tiles.
4. The computer-readable storage medium of claim 1 , wherein performing the blending comprises an alpha-blending approach wherein a weight alpha is determined by raising a normalized distance of a pixel from an edge of its tile to an exponent.
5. The computer-readable storage medium of claim 1 , wherein determining the absolute homographies for each tile comprises solving a linear system such that a global measure of registration is minimized.
6. The computer-readable storage medium of claim 1 , wherein determining the absolute homographies for each tile comprises determining connected components given the pairwise registration of the plurality of tiles.
7. A computer-implemented method of performing image mosaicking, the method performed by the computer comprising the steps of:
performing pairwise registration of a plurality of tiles;
receiving prior homography information for each of the plurality of tiles;
determining absolute homographies for each of the plurality of tiles according to the pairwise registration by minimizing a following objective function
E
(
p
1
,
…
,
p
N
)
=
1
2
∑
i
=
1
N
∑
j
∈
Nbd
(
i
)
p
i
-
p
j
-
p
~
ij
2
+
λ
2
∑
i
=
1
N
p
i
-
p
~
i
2
,
wherein N is a number of tiles, Nbd(i) denote a set of tiles that share any overlap with tile i, p i denotes a homography of tile i, {tilde over (p)} ij denotes a translation obtained by applying the pairwise registration between neighboring tiles i and j, {tilde over (p)} i denotes the prior homography of tile i, and λ denotes a predetermined weighting for the prior homography; and
performing a blending of the plurality of tiles to obtain a stitched image according to the absolute homographies.
8. The method of claim 7 , wherein performing the pairwise registration comprises maximizing a similarity measure.
9. The method of claim 7 , wherein determining the absolute homographies for each of the plurality of tiles comprises revising the pairwise registration according to an objective function for an edge incidence matrix of the plurality of tiles.
10. The method of claim 7 , wherein performing the blending comprises an alpha-blending approach wherein a weight alpha is determined by raising a normalized distance of a pixel from an edge of its tile to an exponent.
11. The method of claim 7 , wherein determining the absolute homographies for each tile comprises solving a linear system such that a global measure of registration is minimized.
12. The method of claim 7 , wherein determining the absolute homographies for each tile comprises determining connected components given the pairwise registration of the plurality of tiles.