Method and system utilizing parameter-less filter for substantially reducing streak and or noise in computer tomography (CT) images
View Patent ↗Photon starvation causes streaks and noise and seriously impairs the diagnostic value of the CT imaging. To reduce streaks and noise, a new scheme of adaptive Gaussian filtering relies on the diffusion-derived scale-space concept in one embodiment of the current invention. In scale-space view, filtering by Gaussians of different sizes is similar to decompose the data into a sequence of scales. As the scale measure, the variance of the filter linearly relates to the noise standard deviation of a predetermined noise model in the new filtering method. The new filter has only one optional parameter that remains stable once tuned. Although single-pass processing using the new filter generally achieves desired results, iterations are optionally performed.
1. A method of equally reducing noise in measured signals, comprising the steps of:
a) determining a relative value in noise variance at each of the measured signals based upon a function of a predetermined noise model to generate a noise-model based variance;
b) automatically generating a discrete filter kernel of a noise-equalizing filter for each of the measured signals based upon the noise-model based variance; and
c) applying the discrete filter kernel to a corresponding one of the measured signals to generate a filtered measured signals.
2. The method of processing measured signals according to claim 1 wherein the noise-model based variance is defined by the measured signal and a known electronic noise value.
3. The method of processing measured signals according to claim 2 wherein the noise-model based variance is
1
+
V
e
/
I
I
,
where the I is the measured signal while V e is the known electronic noise value.
4. The method of processing measured signals according to claim 3 wherein the discrete kernel of the filter for a particular detector element (i) is defined by
I
i
0
G
=
∑
i
I
i
exp
(
-
Δ
x
i
2
2
V
i
0
)
exp
(
-
(
I
i
0
-
I
i
)
2
2
V
R
)
∑
i
exp
(
-
Δ
x
i
2
2
V
i
0
)
exp
(
-
(
I
i
0
-
I
i
)
2
2
V
R
)
where Δx i is the distance of a i th pixel in 1D to a predetermined reference pixel i 0 and Vi 0 is the frequency response of the filter at the reference pixel i 0 , V R is a parameter of the filter.
5. The method of processing measured signals according to claim 3 wherein the discrete kernel of the filter for a particular detector element (i, j) is defined by
I
(
i
0
,
j
0
)
G
=
∑
i
,
j
I
i
,
j
exp
(
-
Δ
x
i
,
j
2
2
V
i
0
,
j
0
)
exp
(
-
(
I
i
0
,
j
0
-
I
i
,
j
)
2
2
V
R
)
∑
i
,
j
exp
(
-
Δ
x
i
,
j
2
2
V
i
0
,
j
0
)
exp
(
-
(
I
i
0
,
j
0
-
I
i
,
j
)
2
2
V
R
)
where Δx i,j is the distance of a (i th , j th ) pixel in 2D to a predetermined reference pixel (i 0 , j 0 ) and Vi 0 , j 0 is the frequency response of the filter at the reference pixel (i 0 , j 0 ), V R is a parameter of the filter.
6. The method of processing measured signals according to claim 2 further comprising an additional step of iterating said steps a), b) and c).
7. The method of processing measured signals according to claim 2 wherein the noise-equalizing filter is a high pass filter including a Gaussian filter.
8. A system for equally reducing noise in measured signals, comprising:
a noise reduction device for performing the tasks of determining a noise variance at each of the measured signals based upon a function of a predetermined noise model to generate a noise-model based variance, said noise reduction device performing the tasks of automatically generating a discrete filter kernel of a noise-equalizing filter for each of the measured signals based upon the noise-model based variance, said noise reduction device performing the tasks of applying the discrete filter kernel to a corresponding one of the measured signals to generate a filtered measured signal.
9. The system for processing measured signals according to claim 8 wherein the noise-model based variance is defined by the measured signal and a known electronic noise value.
10. The system for processing measured signals according to claim 9 wherein the noise-model based variance is
1
+
V
e
/
I
I
,
where the I is the measured signal while V e is the known electronic noise value.
11. The system for processing measured signals according to claim 10 wherein the discrete kernel of the filter for a particular detector element (i) is defined by
I
i
0
G
=
∑
i
I
i
exp
(
-
Δ
x
i
2
2
V
i
0
)
exp
(
-
(
I
i
0
-
I
i
)
2
2
V
R
)
∑
i
exp
(
-
Δ
x
i
2
2
V
i
0
)
exp
(
-
(
I
i
0
-
I
i
)
2
2
V
R
)
where Δx i is the distance of a i th pixel in 1D to a predetermined reference pixel i 0 and Vi 0 is the frequency response of the filter at the reference pixel i 0 , V R is a parameter of the filter.
12. The system for processing measured signals according to claim 10 wherein the discrete kernel of the filter for a particular detector element (i, j) is defined by
I
(
i
0
,
j
0
)
G
=
∑
i
,
j
I
i
,
j
exp
(
-
Δ
x
i
,
j
2
2
V
i
0
,
j
0
)
exp
(
-
(
I
i
0
,
j
0
-
I
i
,
j
)
2
2
V
R
)
∑
i
,
j
exp
(
-
Δ
x
i
,
j
2
2
V
i
0
,
j
0
)
exp
(
-
(
I
i
0
,
j
0
-
I
i
,
j
)
2
2
V
R
)
where Δx i,j is the distance of a (i th , j th ) pixel in 2D to a predetermined reference pixel (i 0 , j 0 ) and Vi 0 , j 0 is the frequency response of the filter at the reference pixel (i 0 , j o ), V R xis a parameter of the filter.
13. The system for processing measured signals according to claim 9 wherein said noise reduction device iterates each of the tasks.
14. The system for processing measured signals according to claim 9 wherein the noise-equalizing filter is a high pass filter including a Gaussian filter.
15. A method of equally reducing noise in measured signals, comprising the steps of:
a) determining a noise variance at each of the measured signals based upon a function of a predetermined noise model;
b) automatically generating a discrete filter kernel of a noise-equalizing filter for each of the measured signals based upon the noise variance; and
c) applying the discrete filter kernel to a corresponding one of the measured signals to generate a filtered measured signals, wherein the noise variance is
1
+
V
e
/
I
I
,
where the I is the measured signal while V e is a known electronic noise value.
16. A system for equally reducing noise in measured signals, comprising:
a noise reduction device for performing the tasks of determining a noise variance at each of the measured signals based upon a function of a predetermined noise model, said noise reduction device performing the tasks of automatically generating a discrete filter kernel of a noise-equalizing filter for each of the measured signals based upon the noise variance, said noise reduction device performing the tasks of applying the discrete filter kernel to a corresponding one of the measured signals to generate a filtered measured signal, wherein the noise variance is
1
+
V
e
/
I
I
,
where the I is the measured signal while Ve is a known electronic noise value.