Method and apparatus for multi-scale based dynamic range compression and noise suppression
View Patent ↗An input image is decomposed into a residual image and a set of detail images in different levels of a hierarchy. The set of detail images are enhanced after which noise reduction filtering is directly applied to the enhanced set of detail images without the need to apply noise estimation to a local area of the enhanced set of detail image in advance. The residual image and the set of detail images with reduced noise are recomposed so as to obtain resultant image.
1. An apparatus for processing an input image, comprising:
an image decomposition unit for decomposing the input image into a residual image and a set of detail images in different levels of a hierarchy, wherein at least one detail image differs from the input image and comprises detail information from a high frequency band;
an image enhancement unit for enhancing the set of detail images;
a noise suppression unit for reducing noise of the enhanced set of detail images without applying noise estimation to a local area of the enhanced set of detail images in advance,
wherein the noise suppression unit directly applies noise reduction processing to each of the enhanced detail images;
an image reconstruction unit for recomposing the residual image and the set of detail images with reduced noise so as to obtain a resultant image; and
a dynamic range compression unit for applying dynamic range compression to the enhanced set of detail images and the residual image;
wherein the dynamic range compression unit compresses the residual image by weighted averaging the residual image and a uniform image.
2. The apparatus according to claim 1 , wherein the noise suppression is a noise filter for retaining detail edge information for the enhanced set of detail images while reducing the noise of the enhanced set of detail images.
3. The apparatus according to claim 2 , wherein, with the noise filter, a pixel that is closer to a current pixel has greater influence on filtering of the current pixel, and a pixel that has a value differing farther from that of the current pixel has less influence on filtering of the current pixel.
4. The apparatus according to claim 3 , wherein the noise filter uses the following equation to reduce the noise of the enhanced set of detail images:
L
″
[
p
,
q
]
=
∑
j
≠
0
,
k
≠
0
L
′
[
p
+
j
,
q
+
k
]
·
exp
(
-
r
2
3
σ
2
-
(
L
′
[
p
+
j
,
q
+
k
]
-
L
′
[
p
,
q
]
)
2
t
2
)
∑
j
≠
0
,
k
≠
0
exp
(
-
r
2
3
σ
2
-
(
L
′
[
p
+
j
,
q
+
k
]
-
L
′
[
p
,
q
]
)
2
t
2
)
where, L′ and L″ are respectively images before and after filtering; r is the distance between pixel (p, q) and pixel (p+j, q+k) in the image L′, p and q are integers greater than or equal to 0, j and k are integers not equal to 0, and the values of j and k depending on the size of the noise filter; for the noise filter with a size of (2x+1)×(2y+1), |j|≦x and |k|≦y, and x and y are positive integers; σ is a constant for controlling the degree of spatial smoothness, the value of which is a real number greater than 0; and t is a constant representing a gray scale threshold which is a positive integer not greater than the maximum gray scale value in the image L′.
5. The apparatus according to claim 1 , wherein the dynamic range compression unit performs the weighted averaging on the residual image and a uniform image according to the following equation:
I″ n-1 [p,q]=I′ n-1 [p,q]*α+B[p,q]*( 1−α)
where I′ n-1 [p,q] and I″ n-1 [p,q] are respectively the residual images before and after compressing; p and q being integers greater than or equal to 0; α is a weighting coefficient comprising a real number between 0 and 1; B[p,q] is a uniform image, wherein the value of each pixel is a constant, the constant calculated based on a mean value of pixels within a region of interest in the input image; the region of interest obtained by performing image segmentation on the input image; and n is a positive integer representing the number of levels of a hierarchy of the input image decomposed by the image decomposition unit.
6. The apparatus according to claim 1 , wherein the dynamic range compression unit further applies a higher compression ratio to compression of contrast for a detail image with higher contrast.
7. The apparatus according to claim 6 , wherein the dynamic range compression unit compresses the detail images according to the following equation:
O i [p,q]=L″ i [p,q]*β i
where, L″ i [p,q] is the detail image before compressing; O i [p,q] is the detail image after compressing; p and q being integers greater than or equal to 0, the value of β i is specified as follows:
β
i
=
{
1
,
i
≤
τ
1
-
1
-
α
n
-
1
-
τ
(
i
-
τ
)
,
i
>
τ
n is a positive integer representing the number of levels of a hierarchy of the input image decomposed by the image decomposition unit; i refers to the level number of the hierarchy in which a compressed detail image is located; τ is a constant, the value of which is an integer between 0 and n; and α is a weighting coefficient comprising a real number between 0 and 1.
8. A method for processing an input image, comprising:
decomposing the input image into a residual image and a set of detail images in different levels of a hierarchy, wherein at least one detail image differs from the input image and comprises detail information from a high frequency band;
enhancing the set of detail images;
reducing noise of the enhanced set of detail images without applying noise estimation to a local area of the enhanced set of detail images in advance,
wherein reducing noise comprises directly applying noise reduction processing to each of the enhanced detail images;
recomposing the residual image and the set of detail images with reduced noise so as to obtain a resultant image; and
applying dynamic range compression to the enhanced set of detail images and the residual image;
wherein the dynamic range compression step compresses the residual image by weighted averaging the residual image and a uniform image.
9. The method according to claim 8 , wherein in the noise reduction step, detail edge information for the enhanced set of detail images is able to be retained while the noise of the enhanced set of detail images is reduced.
10. The method according to claim 9 , wherein in the noise reduction step, a pixel that is closer to a current pixel has greater influence on filtering of the current pixel, and a pixel that has a value differing farther from that of the current pixel has less influence on filtering of the current pixel.
11. The method according to claim 10 , wherein, in the noise reduction step, the following equation is used to reduce the noise of the enhanced set of detail images:
L
″
[
p
,
q
]
=
∑
j
≠
0
,
k
≠
0
L
′
[
p
+
j
,
q
+
k
]
·
exp
(
-
r
2
3
σ
2
-
(
L
′
[
p
+
j
,
q
+
k
]
-
L
′
[
p
,
q
]
)
2
t
2
)
∑
j
≠
0
,
k
≠
0
exp
(
-
r
2
3
σ
2
-
(
L
′
[
p
+
j
,
q
+
k
]
-
L
′
[
p
,
q
]
)
2
t
2
)
where, L′ and L″ are respectively images before and after filtering; r is the distance between pixel (p, q) and pixel (p+j, q+k) in the image L′, p and q being integers greater than or equal to 0, j and k being integers not equal to 0, and the values of j and k depending on the size of a filter for the noise filtering step; for the filter with a size of (2x+1)×(2y+1), |j|≦x and |k|≦y, and x and y are positive integers;
a is a constant for controlling the degree of spatial smoothness, the value of which is a real number greater than 0; and t is a constant representing a gray scale threshold which is a positive integer not greater than the maximum gray scale value in the image L′.
12. The method according to claim 8 , wherein the dynamic range compression step performs the weighted averaging on the residual image and a uniform image according to the following equation:
I″ n-1 [p,q]=I′ n-1 [p,q]*α+B[p,q]*( 1−α)
where I′ n-1 [p,q] and I″ n-1 [p,q] are respectively the residual images before and after compressing; p and q being integers greater than or equal to 0; α is a weighting coefficient comprising a real number between 0 and 1; B[p,q] is a uniform image, wherein the value of each pixel is a constant, the constant calculated based on a mean value of pixels within a region of interest in the input image, the region of interest obtained by performing image segmentation on the input image; and n is a positive integer representing the number of levels of a hierarchy of the input image decomposed by the image decomposition step.
13. The method according to claim 8 , further comprising applying a higher compression ratio to compression of contrast for a detail image with higher contrast.
14. The method according to claim 13 , wherein the detail image is compressed according to the following equation:
O i [p,q]=L″ i [p,q]*β i
where, L″ i [p,q] is the detail image before compressing; O i [p,q] is the detail image after compressing, p and q being integers greater than or equal to 0; and the value of β i is specified as follows:
β
i
=
{
1
,
i
≤
τ
1
-
1
-
α
n
-
1
-
τ
(
i
-
τ
)
,
i
>
τ
n is a positive integer representing the number of levels of a hierarchy of the input image decomposed in the image decomposition step; i refers to the level number of the hierarchy in which a compressed detail image is located; τ is a constant, the value of which is an integer between 0 and n; and α is a weighting coefficient comprising a real number between 0 and 1.
15. An apparatus for processing an input image, comprising:
means for decomposing the input image into a residual image and a set of detail images in different levels of a hierarchy, wherein at least one detail image differs from the input image and comprises detail information from a high frequency band;
means for enhancing the set of detail images;
means for reducing noise of the enhanced set of detail images without applying noise estimation to a local area of the enhanced set of detail images in advance,
wherein the means for reducing noise directly applies noise reduction processing to each of the enhanced detail images;
means for recomposing the residual image and the set of detail images with reduced noise so as to obtain a resultant image; and
means for applying dynamic range compression to the enhanced set of detail images and the residual image;
wherein the means for applying dynamic range compression compresses the residual image by weighted averaging the residual image and a uniform image.
16. A non-transitory computer-readable storage medium comprising program code for causing a computer to perform a method for processing an input image, the method comprising:
decomposing the input image into a residual image and a set of detail images in different levels of a hierarchy, wherein at least one detail image differs from the input image and comprises detail information from a high frequency band;
enhancing the set of detail images;
reducing noise of the enhanced set of detail images without applying noise estimation to a local area of the enhanced set of detail images in advance,
wherein reducing noise comprises directly applying noise reduction processing to each of the enhanced detail images; and
recomposing the residual image and the set of detail images with reduced noise so as to obtain a resultant image; and
applying dynamic range compression to the enhanced set of detail images and the residual image;
wherein the dynamic range compression step compresses the residual image by weighted averaging the residual image and a uniform image.
17. An apparatus for processing an input image, comprising:
an image decomposition unit for decomposing the input image into a residual image and a set of detail images in different levels of a hierarchy, wherein at least one detail image differs from the input image and comprises detail information from a high frequency band;
an image enhancement unit for enhancing the set of detail images;
a noise suppression unit for reducing noise of the enhanced set of detail images without applying noise estimation to a local area of the enhanced set of detail images in advance,
wherein the noise suppression unit directly applies noise reduction processing to each of the enhanced detail images; and
an image reconstruction unit for recomposing the residual image and the set of detail images with reduced noise so as to obtain a resultant image;
wherein the noise suppression is a noise filter for retaining detail edge information for the enhanced set of detail images while reducing the noise of the enhanced set of detail images;
wherein, with the noise filter, a pixel that is closer to a current pixel has greater influence on filtering of the current pixel, and a pixel that has a value differing farther from that of the current pixel has less influence on filtering of the current pixel;
wherein the noise filter uses the following equation to reduce the noise of the enhanced set of detail images:
L
″
[
p
,
q
]
=
∑
j
≠
0
,
k
≠
0
L
′
[
p
+
j
,
q
+
k
]
·
exp
(
-
r
2
3
σ
2
-
(
L
′
[
p
+
j
,
q
+
k
]
-
L
′
[
p
,
q
]
)
2
t
2
)
∑
j
≠
0
,
k
≠
0
exp
(
-
r
2
3
σ
2
-
(
L
′
[
p
+
j
,
q
+
k
]
-
L
′
[
p
,
q
]
)
2
t
2
)
where, L′ and L″ are respectively images before and after filtering; r is the distance between pixel (p, q) and pixel (p+j, q+k) in the image L′, p and q are integers greater than or equal to 0, j and k are integers not equal to 0, and the values of j and k depending on the size of the noise filter; for the noise filter with a size of (2x+1)×(2y+1), |j|≦x and |k|≦y, and x and y are positive integers; σ is a constant for controlling the degree of spatial smoothness, the value of which is a real number greater than 0; and t is a constant representing a gray scale threshold which is a positive integer not greater than the maximum gray scale value in the image L′.