Reducing halo-like effects in motion-compensated interpolation
View Patent ↗The invention relates to a motion-compensated interpolation of a data-signal, which data-signal comprises successive images wherein each image comprises groups of pixels, in which motion vectors are generated ( 18 ), each motion vector corresponding to a group of pixels of one image, between a group of pixels of said one image and a second group of pixels of another image in the data-signal, and interpolated results are obtained ( 16 ) as a function of these motion vectors. In accordance with the present invention, the reliability of each motion vector corresponding to a particular group of pixels is estimated ( 20 ), weights are calculated as a function of the reliability of the motion vectors, and interpolated luminous intensities of groups of pixels are generated for an interpolated image by calculating, on the basis of these weights, weighted averages of the interpolated results.
1. A method of motion-compensated interpolation of a data-signal, said data-signal comprising successive images wherein each image comprises groups of pixels, the method comprising the steps of:
generating motion vectors, each motion vector corresponding to a group of pixels of one image, between a group of pixels of said one image and a second group of pixels of another image in the data-signal;
generating interpolated results as a function of these motion vectors;
estimating the reliability of each motion vector corresponding to a particular group of pixels;
calculating weights as a function of the reliability of the motion vectors; and
generating an interpolated luminous intensity of a group of pixels for an interpolated image by calculating, on the basis of these weights, a weighted average of the interpolated results.
2. The method as claimed in claim 1 , wherein the generation of interpolated luminous intensities is only performed in those parts of the images of the data-signal where edges in the motion vector field of the images are located.
3. A method of motion-compensated interpolation of a data-signal, said data-signal comprising successive images wherein each image comprises groups of pixels, the method comprising the steps of:
generating motion vectors, each motion vector corresponding to a group of pixels of one image, between a group of pixels of said one image and a second group of pixels of another image in the data-signal;
generating interpolated results as a function of these motion vectors;
estimating the reliability of each motion vector corresponding to a particular group of pixels;
calculating weights as a function of the reliability of the motion vectors; and
generating an interpolated luminous intensity of a group of pixels for an interpolated image by calculating, on the basis of these weights, a weighted average of the interpolated results, wherein the interpolated luminous intensity of a group of pixels is calculated according to:
I
k
+
Δ
(
x
→
)
=
(
∑
m
=
1
,
…
,
M
{
w
m
k
(
x
→
)
*
i
m
k
+
Δ
(
x
→
)
}
)
/
∑
m
=
1
,
…
,
M
{
w
m
k
(
x
→
)
}
,
(
I
)
wherein I k+Δ ({right arrow over (x)}) is the interpolated luminous intensity of the group of pixels of an interpolated image F k+Δ , wherein the location of the group of pixels in the image is defined by the integer two-dimensional vector {right arrow over (x)} and where the real value Δ defines the place of the interpolated image F k+Δ in the image sequence F n , n=1,2, . . . , K,K+1, . . . , N, wherein Σ m=l , . . . , M{.} is a summation from 1 to M over its argument {.} and where w k m ({right arrow over (x)}) is a weight corresponding to the m th interpolation result i k+Δ m ({right arrow over (x)}):
i
m
k
+
Δ
(
x
→
)
=
median
{
(
I
k
(
round
{
x
→
-
Δ
*
D
→
m
k
(
x
→
)
}
)
,
(
I
k
(
x
→
)
+
I
k
+
1
(
x
→
)
)
/
2
)
,
(
I
k
+
1
(
round
{
x
→
+
(
1
-
Δ
)
*
D
→
m
k
(
x
→
)
}
)
}
,
(
II
)
wherein median{.} is a function which gives the median value of its input arguments and round{.} is a function which gives the nearest integer value to each component of its input argument, and wherein I k ({right arrow over (x)}) is a luminous intensity of the group of pixels at location {right arrow over (x)} of the image F k and wherein {right arrow over (D)} m k ({right arrow over (x)}) is the m th two-dimensional integer motion vector, which is normalized between two successive images, of the M motion vectors which correspond to the group of pixels at location {right arrow over (x)} and wherein the weight w k m ({right arrow over (x)}) is a function of the reliability of the motion vector {right arrow over (D)} m k ({right arrow over (x)}).
4. The method as claimed in claim 3 , wherein the reliability of the motion vector {right arrow over (D)} m k ({right arrow over (x)}) is a function of the difference between the luminous intensities I k ({right arrow over (x)}) and I k+1 ({right arrow over (x)}+{right arrow over (D)} m k ({right arrow over (x)})) and wherein the reliability is also a function of the relative frequency of occurrence of {right arrow over (D)} m k ({right arrow over (x)}) in the neighborhood of the location {right arrow over (x)} in the image F k .
5. A method of motion-compensated interpolation of a data-signal, said data-signal comprising successive images wherein each image comprises groups of pixels, the method comprising the steps of:
generating motion vectors, each motion vector corresponding to a group of pixels of one image, between a group of pixels of said one image and a second group of pixels of another image in the data-signal;
generating interpolated results as a function of these motion vectors;
estimating the reliability of each motion vector corresponding to a particular group of pixels;
calculating weights as a function of the reliability of the motion vectors; and
generating an interpolated luminous intensity of a group of pixels for an interpolated image by calculating, on the basis of these weights, a weighted average of the interpolated results, wherein the generation of interpolated luminous intensities is only performed in those parts of the images of the data-signal where edges in the motion vector field of the images are located, and
wherein the method comprises a step of edge detection, wherein an edge in the motion vector field of image F k is detected if at least one of the inequalities (C1) and (C2) is satisfied:
[
D
→
q
k
(
x
→
-
K
→
)
]
1
-
[
D
→
q
k
(
x
→
+
K
→
)
]
1
>
T
,
(
III
)
[
D
→
q
k
(
x
→
-
K
→
)
]
2
-
[
D
→
q
k
(
x
→
+
K
→
)
]
2
>
T
,
(
IV
)
where q is a pre-determined integer value and wherein ∥.∥ is a function which yields the absolute value of its input argument, [.] p is a function which yields the p th component of its vector input argument, where T is a pre-determined fixed real value threshold and wherein {right arrow over (K)} is a vector which is given with:
{right arrow over (K)} =( K 1 ;K 2 ) T , (D)
where K 1 and K 2 are integer values.
6. A device for motion-compensated interpolation of a data-signal, said data-signal comprising successive images wherein each image comprises groups of pixels, the device comprising:
means for generating motion vectors, each motion vector corresponding to a group of pixels of one image, between a group of pixels of said one image and a second group of pixels of another image in the data-signal;
means for generating interpolated results as a function of these motion vectors;
means for estimating the reliability of each motion vector corresponding to a particular group of pixels;
means for calculating weights as a function of the reliability of the motion vectors; and
means for generating interpolated luminous intensities of groups of pixels by calculating, on the basis of these weights, weighted averages of the interpolated results.
7. A picture signal display apparatus, comprising:
means for receiving a data-signal, which data-signal comprises successive images wherein each image comprises groups of pixels;
a device for motion-compensated interpolation of said data-signal, as claimed in claim 6 ;
means for generating at least one interpolated image on the basis of said interpolated luminous intensities; and
means for displaying the at least one interpolated image.