Image denoising device
View Patent ↗An image processing device comprises a store ( 110 ) for receiving image data ( 8 ) with a pixel identifier and at least one piece of associated intensity data and a pilot ( 100 ) which is provided to call a reduction unit ( 120 ) with pixels to be weighted and, for each pixel, in order to define a first neighborhood and a plurality of second neighborhoods with a neighborhood selection unit ( 130 ), in order to obtain a weighting value for the second neighborhood with a weighting unit ( 140 ) with the first neighborhood and a neighborhood among the plurality of second neighborhoods, in order to define each time a piece of weighted intensity data of a working pixel as the weighting value of the second neighborhood multiplied by the intensity data of the associated working pixel, and in order to define a piece of weighted intensity data of the pixel to be weighted by adding each of the pieces of weighted intensity data of a working pixel. The weighting unit ( 140 ) is provided to calculate the distance between first and second neighborhoods on the basis of a sum, where each square of the difference between the respective intensity data of these neighborhoods is modulated by a quantity which is inversely proportional to the intensity data in the second neighborhood.
1. An image processing device comprising:
a storage unit for receiving image data, each image data comprising a pixel identifier and associated intensity data,
a reduction unit configured to select working pixels around a given pixel on the basis of a reduction rule,
a neighbourhood selection unit configured to select a neighbourhood of a given pixel on the basis of a neighbourhood rule,
a weighting unit configured to calculate a weighting value between two neighbourhoods on the basis of distances between the intensity data in said two neighbourhoods and
a pilot unit configured to call the reduction unit with pixels to be weighted and for each pixel to be weighted, and further configured to:
call the neighbourhood selection unit with the pixel to be weighted to define a first neighbourhood and with the associated working pixels to define a plurality of second neighbourhoods,
repeatedly call the weighting unit with the first neighbourhood and a second neighbourhood from the plurality of second neighbourhoods to obtain, each time, a weighting value for the second neighbourhood,
multiply each weighting value of the second neighbourhood by the intensity data of the associated working pixel to define, each time, a piece of weighted intensity data of a working pixel,
wherein the weighting unit is further configured to associate the values with a neighbourhood in a vector, and to define a distance d between a first vector u(Ni) and a second vector u(Ni) on the basis of a sum of quotients, wherein
a numerator of the distance d depends on the square of the difference between values taken from respective columns of the first and second vectors, and
a denominator of the distance d depends on a quantity which is inversely proportional to a value associated with the respective column of the second vector, and
wherein the distance d raised to the square d(u(Ni)-u(Nj)) 2 between two vectors u(Ni) and u(Nj) is calculated on the basis of the relationship:
d
p
(
u
(
N
i
)
-
u
(
N
j
)
)
2
=
∑
p
=
1
p
=
p
(
u
(
p
)
(
N
i
)
-
u
(
p
)
(
N
j
)
)
2
(
u
(
p
)
(
N
i
)
-
u
(
p
)
(
N
j
)
)
2
γ
2
,
(
51
)
where P is the number of columns of the vector u(N), u (p) (N) is the index column p of the vector u (N) and where γ is a non-zero coefficient.
2. The image processing device according to claim 1 , wherein the pilot unit is further configured to call pixels to be weighted in accordance with a section rule.
3. The image processing device according to claim 2 , wherein the pilot unit is further configured to calculate a weighted intensity value for each pixel in the first neighbourhood which is associated with a selected pixel to be weighted, on the basis of the weighting values associated with the pixel to be weighted.
4. The image processing device according to claim 3 , wherein the pilot unit is further configured to define the weighted intensity value of a pixel belonging to a plurality of first separate neighbourhoods as the mean of the weighted intensity values associated with that pixel in each of said plurality of separate neighbourhoods.