Methods and systems for image alignment of at least one image to a model
A system and a method for image alignment between at least two images to a three-dimensional model. The method including: determining a lower bound and an upper bound of an acceptable likelihood of mismatch between the at least two images; evaluating the likelihood of mismatch between the at least two images over a set of poses (r), shifts (t), or both poses (r) and shifts (t); and discarding those evaluations resulting beyond the lower bound and upper bound.
1. A method for image alignment of at least one two-dimensional or three-dimensional image to a two-dimensional or three-dimensional model, executed on a processing unit, the image alignment having an acceptable likelihood of mismatch between the at least one image and the model, the method comprising:
selecting a value for a radius in Fourier space;
discretizing a set of poses into a discrete grid of candidate poses and a set of shifts into a discrete grid of candidate shifts;
determining a fixed fraction as an upper bound on the acceptable likelihood of mismatch, the fixed fraction being determined based on, at least, the fraction of poses and shifts that are discarded on typical dataset of images;
the processing unit iteratively determining whether a selected accuracy of image alignment has been obtained, and when such determination is false:
assigning a lower bound to the acceptable likelihood of mismatch, the lower bound comprising a first component;
analyzing the image to isolate selected portions of the image that are below the value for a radius in Fourier space;
determining values for the first component for each of the poses and shifts on the discrete grid of candidate poses and the discrete grid of shifts, using only the isolated selected portions of the image;
determining a reference first component using a value for the fixed fraction;
parsing the discrete grid of candidate poses by analyzing each one of the candidate poses over all of the candidate shifts to obtain a minimum value of the first component over all the candidate shifts and discarding each candidate pose from the discrete grid if the first component exceeds the reference first component;
parsing the discrete grid of candidate shifts by analyzing each of the candidate shifts over all of the candidate poses to obtain a minimum value of the first component over all candidate poses and discarding each candidate shift from the discrete grid if the first component exceeds the reference first component;
for every remaining pose in the discrete grid of candidate poses, replacing the pose with a plurality of subdivided grid points representing the candidate poses;
for every remaining shift in the discrete grid of candidate shifts, replacing the shift with a plurality of subdivided grid points representing the candidate shifts; and
increasing the radius in the Fourier space; and
otherwise, returning the pose and shift at the lower bound with minimum value.
2. The method of claim 1 , wherein the lower bound is determined with the images at a resolution that is less than the maximum resolution for the images.
3. The method of claim 1 , wherein the first component is the squared error of Fourier coefficients at or below a selected radius in Fourier space, and wherein the lower bound further comprises a second component that is the squared error of Fourier coefficients above the selected radius.
4. The method of claim 1 , wherein the lower bound further comprises a second component that comprises:
V
1
-
∑
l
>
L
1
2
C
l
2
Y
^
l
2
-
4
∑
l
>
L
1
2
C
l
2
Y
^
l
2
,
wherein V 1 is the power of one of the images at high frequencies, subscript l denotes a wavevector, subscript L denotes the selected radius in the Fourier space, C is a contrast transfer function (CTF) of the image-capturing apparatus, and Y is a vector representing a projection of the three-dimensional model.
5. The method of claim 4 , wherein the second component is only recomputed if the CTF of the image-capturing apparatus is different.
6. The method of claim 1 , wherein the determination of the upper bound comprises evaluating a value for the likelihood of mismatch at a specific pose, specific shift, or both.
7. The method of claim 1 , wherein determining the reference first component (U*) comprises determining the reference first component (U*) such that:
{
(
r
,
t
)
;
U
(
r
,
t
)
≤
U
*
}
{
(
r
,
t
)
;
U
(
r
,
t
)
>
U
*
}
=
f
keep
,
wherein f keep is the fixed fraction.
8. The method of claim 1 , wherein the at least one image is two-dimensional and the model is three-dimensional, wherein replacing the pose with the plurality of subdivided grid points comprises replacing the pose with eight subdivided grid points, and wherein replacing the shift with the plurality of subdivided grid points comprises replacing the shift with four subdivided grid points.
9. The method of claim 1 , wherein the at least one image is two-dimensional and the model is two-dimensional, wherein replacing the pose with the plurality of subdivided grid points comprises replacing the pose with two subdivided grid points, and wherein replacing the shift with the plurality of subdivided grid points comprises replacing the shift with four subdivided grid points.
10. The method of claim 1 , wherein the at least one image is three-dimensional and the model is three-dimensional, wherein replacing the pose with the plurality of subdivided grid points comprises replacing the pose with eight subdivided grid points, and wherein replacing the shift with the plurality of subdivided grid points comprises replacing the shift with eight subdivided grid points.
11. The method of claim 1 , wherein the selected accuracy of image alignment is obtained when the value for the radius in Fourier space is equal to the Nyquist rate.
12. The method of claim 1 , wherein the selected accuracy of image alignment is obtained when a selected number of iterations have been performed.
13. The method of claim 1 , wherein the fixed fraction is between 3% and 10%.
14. The method of claim 1 , wherein the fixed fraction is approximately 5%.