Feature value extracting device, object identification device, and feature value extracting method
View Patent ↗A feature calculator calculates the number of edge pixel pairs (three-dimensional histogram) defined by the edge direction of a predetermined pixel, the edge direction of an edge pixel present in a neighboring area of the predetermined pixel, and the spatial position relation between the predetermined pixel and the edge pixel present in the neighboring area of the predetermined pixel as a feature of an image. With this, the feature also represents the spatial relationship of the edges. Consequently, it is possible to also identify the difference of edge patterns due to the difference of the spatial relationships of the edges and increase the accuracy of identification of, for example, a portrait image.
1. A feature extracting apparatus comprising:
an edge image extractor that extracts edges from an image; and
a feature calculator that calculates a spatial co-occurrence matrix as a feature of the image, the spatial co-occurrence matrix including elements for a number of combinations between at least one of an edge direction of a predetermined pixel and an edge direction of an edge pixel present in a neighboring area of the predetermined pixel, and a position of a division area including the edge pixel in the neighboring area of division areas resulting from dividing the neighboring area.
2. The feature extracting apparatus according to claim 1 , wherein the spatial co-occurrence matrix includes elements for a number of products of an edge direction number of the predetermined pixel, an edge direction number of an edge pixel present in each division area in the neighboring area and a number of division areas.
3. A feature extracting apparatus comprising:
an edge image extractor that extracts edges from an image; and
a feature calculator that calculates a feature of the image by equation 1 or equation 2 below:
H
1
(
i
,
k
)
=
1
NM
∑
x
=
0
M
-
1
∑
y
=
0
N
-
1
[
E
i
(
x
,
y
)
∑
j
=
1
N
i
∑
(
x
′
,
y
′
)
∈
R
k
(
x
,
y
)
E
j
(
x
′
,
y
′
)
]
(
Equation
1
)
i, j: Edge direction number, i=1, . . . , N i , j=1, . . . , N i
N i : Total number of edge directions
k: Edge pixel position number, k=1, . . . , N k
N k : Total number of division areas
H
2
(
j
,
k
)
=
∑
i
=
1
N
j
1
NM
∑
x
=
0
M
-
1
∑
y
=
0
N
-
1
[
E
i
(
x
,
y
)
∑
(
x
′
,
y
′
)
∈
R
k
(
x
,
y
)
E
j
(
x
′
,
y
′
)
]
(
Equation
2
)
i, j: Edge direction number, i=1, . . . , N j , j=1, . . . , N j
N j : Total number of edge directions
k: Edge pixel position number, k=1, . . . , N k
N k : Total number of division areas
where in equation 1 and equation 2, N and M indicate a vertical-direction and horizontal-direction number of pixels of an object image for which the feature is calculated, (x, y) indicates coordinates of a predetermined pixel, R k (x, y) indicates a k′th division area in a neighboring area of pixel (x, y), and E i and E j are defined by the following equation 3:
E i ( x,y )=1, if the edge direction at position ( x,y ) is i.
E i ( x,y )=0, otherwise (Equation 3)
i: Edge direction number i=1, . . . , N i
Ni: Total number of edge directions.
4. A feature extracting apparatus comprising:
an edge image extractor that extracts edges from an image; and
a feature calculator that calculates a feature of the image by means of equation 4 below:
H
sco
(
i
,
j
,
k
)
=
1
NM
∑
x
=
0
M
-
1
∑
y
=
0
N
-
1
[
E
i
(
x
,
y
)
∑
(
x
′
,
y
′
)
∈
R
k
(
x
,
y
)
E
j
(
x
′
,
y
′
)
]
(
Equation
4
)
i, j: Edge direction number, i=1, . . . , N i , j=1, . . . , N i
N i : Total number of edge directions
k: Edge pixel position number, k=1, . . . , N k
N k : Total number of division areas
where, in equation 4, N and M indicate a vertical-direction and horizontal-direction number of pixels of an object image for which a feature is calculated, (x, y) indicates coordinates of a predetermined pixel, R k (x, y) indicates a k′th division area in a neighboring area of pixel (x, y), and E i and E j are defined by equation 5 below:
E i ( x,y )=1, if the edge direction at position ( x,y ) is i .
E i ( x,y )=0, otherwise (Equation 5)
i: Edge direction number i=1, . . . , N i
Ni: Total number of edge directions.
5. An object identification apparatus comprising:
an edge extractor that generate edge images from a grayscale scale image; and
a feature calculator that calculates a number of edge pixels stipulated by at least one of an edge direction of a predetermined pixel and an edge direction of an edge pixel present in a neighboring area of the predetermined pixel, and a spatial relationship between the predetermined pixel and the edge pixel present in the neighboring area, as a feature of an image,
a plurality of weak classifiers that receive a plurality of features obtained by the feature calculator as an input, and output estimates from the input plurality of features and a classification function acquired by learning beforehand;
a combiner that sums the estimates output from the plurality of weak classifiers; and
a determiner that performs a threshold value determination of a summed estimate.
6. A feature extracting method comprising:
extractings edges from an image, by an edge image extractor; and
calculating a spatial co-occurrence matrix as a feature of the image, by a feature calculator, the spatial co-occurrence matrix including elements for a number of combinations between at least one of an edge direction of a predetermined pixel and an edge direction of an edge pixel present in a neigh area of the predetermined pixel, and a position of a division area including the edge pixel in the neighboring area of division areas resulting from dividing the neighboring area.