Apparatus and method for extracting target, and recording medium storing program for performing the method
View Patent ↗A target extracting apparatus and method for extracting a target through probability modeling of pixel brightness values in an image, and a recording medium storing a program for performing the method, are disclosed. The method includes extracting a background image from an input image, modeling brightness values of the input image and the background image based on a Gaussian distribution to generate an input image modeling result and a background image modeling result, calculating likelihood ratios for determining whether a pixel in the input image corresponds to a target based on a correlation between the input image modeling result and the background image modeling result and target templates having different sizes; determining a location of the target based on the calculated likelihood ratios; and outputting the determined location of the target.
1. A method of extracting a target, the method comprising:
extracting a background image from an input image;
modeling brightness values of the input image and the background image based on a Gaussian distribution to generate an input image modeling result and a background image modeling result;
calculating likelihood ratios for determining whether a pixel in the input image corresponds to a target based on a correlation between the input image modeling result and the background image modeling result and target templates having different sizes;
determining a location of the target based on the calculated likelihood ratios; and
outputting the determined location of the target,
wherein, for a plurality of targets, the determining of the locations of the targets comprises:
receiving the likelihood ratios and a threshold for target determination;
generating a target map having a same size as the input image, setting areas on the target map to have a first value when each of the likelihood ratios for the areas is greater than the threshold, and setting areas on the target map to have a second value when each of the likelihood ratios for the areas is not greater than the threshold; and
determining each area set to the first value as a single target, setting a representative location for each of the plurality of targets, and determining the representative locations as locations of the plurality of targets.
2. The method of claim 1 , wherein the modeling of the brightness values comprises:
setting a window having a predetermined size about a pixel in each of the input image and the background image;
calculating a mean brightness value of pixels included in the set window;
subtracting the mean brightness value from a brightness value of a center pixel in the set window; and
generating an image in which the mean brightness value is subtracted from the brightness value of the center pixel in the set window while scanning the input image and the background image with the set window.
3. The method of claim 1 , wherein the calculating of the likelihood ratios comprises:
receiving Gaussian distribution modeling results of the input image and the background image;
receiving the target templates having different sizes;
calculating a correlation between the input image modeling result and the background image modeling result; and
acquiring likelihood ratios for each of a plurality of pixels in the input image, a number of the acquired likelihood ratios corresponding to a number of the target templates, and determining a maximum likelihood ratio from among the acquired likelihood ratios as a likelihood ratio of the pixel corresponding to the target.
4. The method of claim 3 , wherein in the receiving of the target templates having different sizes, if the target templates having different sizes are set in advance, the target templates use brightness distribution values of the set target templates.
5. The method of claim 3 , wherein in the receiving of the target templates having different sizes, if the target templates having different sizes are not set in advance, the target templates are set to have a third value.
6. The method of claim 3 , wherein the calculating of the correlation comprises:
extracting a first sub-image having a corresponding target template size from the Gaussian distribution modeling result of the input image;
extracting a second sub-image having the corresponding target template size from the Gaussian distribution modeling result of the background image; and
calculating a correlation indicating a similarity between the input image and the background image by using the first sub-image and the second sub-image.
7. The method of claim 6 , wherein the correlation is calculated by using the equation:
ρ
k
=
∑
i
,
j
Y
k
′
(
i
,
j
)
-
X
k
′
(
i
,
j
)
∑
i
,
j
{
X
k
′
(
i
,
j
)
}
2
,
where Y′ k =Y k − Y , X′ k =X k − X , Y k − Y denotes a first sub-image having a kth target template size, which is extracted from the Gaussian distribution modeling result of the input image,
where i and j are location coordinates of a pixel in the Gaussian distribution modeling result of the input image, Y k is an input image and Y is an average of the input images,
and X k − X denotes a second sub-image having the kth target template size, which is extracted from the Gaussian distribution modeling result of the background image,
where X k is a background image and X is an average of the background images.
8. The method of claim 3 , wherein each of the likelihood ratios is calculated by using the equation:
Λ
0
(
p
)
=
max
k
(
S
k
)
T
[
Y
k
′
-
ρ
k
·
X
k
′
]
/
S
k
,
where
k
∈
{
1
,
…
,
L
}
,
where S k denotes a kth target template from among the target templates having different sizes, Y′ k =Y k − Y , X′ k =X k − X , Y k − Y denotes a first sub-image having a kth target template size based on the Gaussian distribution modeling result of the input image,
p denotes a location of a pixel, T denotes a real number, L denotes a number of the target templates,
Y k is an input image and Y is an average of the input images,
X k − X denotes a second sub-image having the kth target template size based on the Gaussian distribution modeling result of the background image,
where X k is a background image and X is an average of the background images,
and ρ k denotes a correlation calculated from the first sub-image and the second sub-image.
9. The method of claim 1 , wherein, for a single target, the determining of the location of the target comprises determining a location of a pixel having a maximum likelihood ratio from among the calculated likelihood ratios as the location of the target.
10. An apparatus which extracts a target, the apparatus comprising:
a background image extractor which extracts a background image from an input image;
a Gaussian distribution modeler which models brightness values of the input image and the background image based on a Gaussian distribution to generate an input image modeling result and a background image modeling result;
a likelihood ratio calculator which calculates likelihood ratios used to determine whether a pixel in the input image corresponds to a target based on a correlation between the input image modeling result and the background image modeling result and target templates having different sizes; and
a target location determiner which determines a location of the target based on the calculated likelihood ratios and outputs the determined location of the target,
wherein, for a plurality of targets, the target location determiner receives the likelihood ratios and a threshold for target determination, generates a target map having a same size as the input image, sets areas on the target map to have a first value when each of the likelihood ratios for the areas are is greater than the threshold, sets areas on the target map to have a second value when each of the likelihood ratios for the areas is not greater than the threshold, determines each area set to the first value as a single target, sets a representative location for each of the plurality of targets, and determines the representative locations as locations of the plurality of targets.
11. The apparatus of claim 10 , wherein the Gaussian distribution modeler comprises:
a window setter which sets a window having a predetermined size about a pixel in each of the input image and the background image;
a first calculator which calculates a mean brightness value of pixels included in the set window;
a second calculator which subtracts the mean brightness value from a brightness value of a center pixel in the set window; and
a Gaussian distribution image generator which generates an image in which the mean brightness value is subtracted from the brightness value of the center pixel in the set window while scanning the input image and the background image with the set window.
12. The apparatus of claim 10 , wherein the likelihood ratio calculator comprises:
a receiver which receives Gaussian distribution modeling results of the input image and the background image and receives the target templates having different sizes;
a correlation calculator which calculates a correlation between the input image modeling result and the background image modeling result; and
a likelihood ratio determiner which acquires likelihood ratios for each of a plurality of pixels in the input image, a number of the acquired likelihood ratios corresponding to a number of the target templates, and determines a maximum likelihood ratio from among the acquired likelihood ratios as a likelihood ratio of the pixel corresponding to the target.
13. The apparatus of claim 12 , wherein in the receiver, if the target templates having different sizes are set in advance, the target templates use brightness distribution values of the set target templates.
14. The apparatus of claim 12 , wherein in the receiver, if the target templates having different sizes are not set in advance, the target templates are set to have a third value.
15. The apparatus of claim 12 , wherein the correlation calculator calculates the correlation by using the equation:
ρ
k
=
∑
i
,
j
Y
k
′
(
i
,
j
)
-
X
k
′
(
i
,
j
)
∑
i
,
j
{
X
k
′
(
i
,
j
)
}
2
,
where Y′ k =Y k − Y , X′ k =X k − X , Y k − Y denotes a first sub-image having a kth target template size, which is extracted from the Gaussian distribution modeling result of the input image,
where i and j are location coordinates of a pixel in the Gaussian distribution modeling result of the input image, Y k is an input image and Y is an average of the input images,
and X k − X denotes a second sub-image having the kth target template size, which is extracted from the Gaussian distribution modeling result of the background image,
where X k is a background image and X is an average of the background images.
16. The apparatus of claim 12 , wherein the likelihood ratio determiner calculates the likelihood ratios by using the equation:
Λ
0
(
p
)
=
max
k
(
S
k
)
T
[
Y
k
′
-
ρ
k
·
X
k
′
]
/
S
k
,
where
k
∈
{
1
,
…
,
L
}
,
where S k denotes a kth target template from among the target templates having different sizes, Y′ k =Y k − Y , X′ k =X k − X , Y k − Y denotes a first sub-image having a kth target template size based on the Gaussian distribution modeling result of the input image,
p denotes a location of a pixel, T denotes a real number, L denotes a number of the target templates,
Y k is an input image and Y is an average of the input images,
X k − X denotes a second sub-image having the kth target template size based on the Gaussian distribution modeling result of the background image,
where X k is a background image and X is an average of the background images,
and ρ k denotes a correlation calculated from the first sub-image and the second sub-image.
17. The apparatus of claim 10 , wherein, for a single target, the target location determiner determines a location of a pixel having a maximum likelihood ratio from among the calculated likelihood ratios as a location of the target.
18. A non-transitory computer-readable recording medium storing a computer-readable program for executing a method of extracting a target, the method including:
extracting a background image from an input image;
modeling brightness values of the input image and the background image based on a Gaussian distribution to generate an input image modeling result and a background image modeling result;
calculating likelihood ratios for determining whether a pixel in the input image corresponds to a target based on a correlation between the input image modeling result and the background image modeling result and target templates having different sizes;
determining a location of the target based on the calculated likelihood ratios; and
outputting the determined location of the target,
wherein, for a plurality of targets, the determining of the locations of the targets comprises:
receiving the likelihood ratios and a threshold for target determination;
generating a target map having a same size as the input image, setting areas on the target map to have a first value when each of the likelihood ratios for the areas is greater than the threshold, and setting areas on the target map to have a second value when each of the likelihood ratios for the areas is not greater than the threshold; and
determining each area set to the first value as a single target, setting a representative location for each of the plurality of targets, and determining the representative locations as locations of the plurality of targets.