US 2010/0135574 A1 · Image processing using neural network

Abstract

Image processing method wherein each image is composed of an array of image points, so called pixels or voxels particularly in a two-, three-, or more dimensional space respectively each image point being univocally defined by its position within the array of image points and by one or more numerical parameters defining the image point appearance as regards characteristics of brightness, grey, colour shade or the like, and wherein each image point is considered to be a node of an artificial neural network, the image being processed as a function of parameters defining the appearance of each pixel as values of the nodes of said artificial neural network and as a function of connections of each pixel under processing with neighbouring pixels composed of pixels of a predetermined subset of pixels, particularly with neighbouring pixels of said pixel under processing, so called pixel window, while pixels of the new image i.e. of the processed image are obtained by iterative evolution steps of parameters defining the appearance such as evolution steps of the value of nodes or by iterative evolution steps of values of the set of connections or by a combination of said evolutions, wherein the processing occurs by evolution iterative steps where each step is a function also of connections of neighbouring pixels with the pixel under examination, when each of said neighbouring pixels of the pixel under examination is considered also as a neighbouring pixel of one ore more or all pixels adjacent to said neighbouring pixel, which function is an immediate feedback contribution for determining appearance values of all other pixels.

Claims14

Dependent claims

while pixels of the new image i.e. of the processed image are obtained by iterative evolution steps of parameters defining the appearance such as evolution steps of the value of nodes or by iterative evolution steps of values of the set of connections or by a combination of said evolutions,
characterized in that the processing occurs by evolution iterative steps where each step is a function also of connections of neighbouring pixels with the pixel under examination, when each of said neighbouring pixels of the pixel under examination is considered also as a neighbouring pixel of one ore more or all pixels adjacent to said neighbouring pixel, which function is an immediate feedback contribution for determining appearance values of all other pixels in that the appearance valueo of each pixel of the image contributes to the determination of the new appearance values of all the other pixels of the new image resulting from the said interative step.

Method according to one or more of the preceding claims, characterized in that whole processing steps determining an evolution of the value of connections of each image point or pixel or voxel of the given image is composed of the following equations where the following definitions are valid:

Wi, j[0]=0.0;
initial value of the connection between any two pixels.
ui∈[−1+α,1+α];
Initial scaling of the value of all pixels of the given image.
α=scaling threshold of pixel values.
x[n]=x at n step;
any variable (connection, pixel, etc.) at a certain processing cycle.
Pi[n]=i-th Pixel at n step;
Outi[n]=i-th Output at n step;
Si[n]=i-th State at n step;
N=neighbouring nodes.
and such whole can be divided in four steps defined by four equations:
D
i
=
∑
j
N

(
u
j
[
n
]
-
W
i
,
j
[
n
]
)
;
(
Eq


1
)
where the factor Di is merely a sum of differences between the value of each image point, i.e. each pixel and the weight with which the central image point or pixel is connected thereto.
J
i
=

D
i
-

-
D
i

D
i
+

-
D
i
;
(
Eq


2
)
where the variable Ji is the result of the application of the hyperbolic tangent function to the factor Di for determining the following effects:
enclosing possible values of Di in a finite interval;
providing a smooth sigmoid shape to possible values of Di;
working with a function differentiable in each point.
ΔWi, j[n]=−(ui[n]·Ji)·(−2·Ji)·(1−Ji2)·(uj[n]−Wi, j[n]);
(Eq 3)
comprising three components that are the following:
a) the first component, −(ui[n]·Ji), weighting the value of the central image point or central pixel with the variable Ji depending on its neighbourhood (weights and pixels), while it reverses the classical mathematic relation among signs, particularly for example the concordance produces a negative value, while the discordance produces a positive value:
b. the second component of the equation (3), (−2·Ji)·(1−Ji2), is the second derivative of Ji analysing how the variation between neighbouring image points or pixels and weights with which the central Image point or central pixel is connected thereto changes;
c. the third component of the equation (3) weights the whole equation by the difference between each neighbouring image point or pixel and the weight with which the central image point or central pixel is connected thereto.
Wi, j[n+1]=Wi, j[n]+ΔWi, j[n];
(Eq 4)
where weights connecting the central image point or central pixel to each image point or pixel of its neighbourhood are updated.

78 more elements

Dependencies are inferred from the claim text.

Image processing using neural network

Life of this patent

Image processing using neural network

Life of this patent

Abstract

Claims14