Method of automatically performing pet eye correction
View Patent ↗A method of performing eye correction is performed in an electronic device having an image processor, an image sensor, and a storage device. The image sensor captures image data, detects a pet face in the image data, and locates a plurality of candidate eye regions in the pet face. A classifier of the image processor verifies at least one eye region of the plurality of candidate eye regions, and the image processor recovers an abnormal pupil region of the at least one verified eye region.
1. A method of performing eye correction in an electronic device comprising an image processor, an image sensor, and a storage device, the method comprising:
the image sensor capturing image data;
the image processor detecting a pet face in the image data;
the image processor locating a plurality of candidate eye regions in the pet face;
a classifier of the image processor verifying at least one eye region of the plurality of candidate eye regions; and
the image processor recovering an abnormal pupil region of the at least one verified eye region by changing a color of each pixel in the abnormal pupil region to a dark color according to following formulas:
Y new =40 ×B ( f p )+ Y old ×(1− B ( f p ));
Cb new =Cb old ×(1 −B ( f p )); and
Cr new =Cr old ×(1 −B ( f p ));
where
B
(
f
p
)
=
P
(
f
p
|
y
=
1
)
P
(
y
=
1
)
P
(
f
p
|
y
=
1
)
P
(
y
=
1
)
+
P
(
f
p
|
y
=
0
)
P
(
y
=
0
)
,
f
p
is color intensity of the pixel, P(f p |y=1) is probability of the color intensity equaling f p when the pixel is abnormally-colored, P(y=1) is probability of the pixel being abnormally-colored, P(f p |y=0) is probability of the color intensity equaling f p when the pixel is not abnormally-colored, P(y=0) is probability of the pixel not being abnormally-colored, Y old is Y-component of the color, Y new is Y-component of the dark color, Cb old is Cb-component of the color, Cb new is Cb-component of the dark color, Cr new is Cr-component of the color, and Cr old is Cr-component of the dark color.
2. The method of claim 1 , further comprising:
the image processor rejecting at least one non-eye region of the plurality of candidate eye regions.
3. The method of claim 1 , further comprising:
the image processor removing a similar eye region of the at least one verified eye region.
4. The method of claim 1 , wherein the image processor locating the plurality of candidate eye regions in the pet face comprises:
the image processor detecting a center position of an eye region of the plurality of candidate eye regions;
the image processor calculating a radius of the eye region;
the image processor correcting the center position; and
the image processor confirming an abnormal pupil region of the eye region.
5. A method of performing eye correction in an electronic device comprising an image processor, an image sensor, and a storage device, the method comprising:
the image sensor capturing image data;
the image processor detecting a pet face in the image data
the image processor locating a plurality of candidate eye regions in the pet face by:
detecting a center position of an eye region of the plurality of candidate eye regions;
calculating a radius of the eye region;
correcting the center position; and
calculating filling and roundness of a candidate pupil region of the eye region according to following formulas:
Filling
=
S
(
K
1
)
S
(
K
2
)
;
and
Roundness
=
4
×
π
×
S
(
K
2
)
L
2
(
K
2
)
;
where K 1 is a center empty segment, K 2 is the candidate pupil region, S(K 1 ) and S(K 2 ) represent areas of the first segment K 1 and K 2 , respectively, and L(K 2 ) is perimeter of K 2 ; and
the image processor confirming the candidate abnormal region when Filling<0.75 and Roundness<0.9;
a classifier of the image processor verifying at least one eye region of the plurality of candidate eye regions; and
the image processor recovering an abnormal pupil region of the at least one verified eye region.
6. A method of performing eye correction in an electronic device comprising an image processor, an image sensor, and a storage device, the method comprising:
the image sensor capturing image data;
the image processor detecting abnormally-colored regions in the image data;
the image processor locating a plurality of candidate eye regions according to the abnormally-colored regions;
a classifier of the image processor verifying at least one eye region of the plurality of candidate eye regions; and
the image processor recovering an abnormal pupil region of the at least one verified eye region by changing a color of each pixel in the abnormal pupil region to a dark color according to following formulas:
Y new =40 ×B ( f p )+ Y old ×(1 −B ( f p ));
Cb new =Cb old ×(1 −B ( f p ));and
Cr new =Cr old ×(1 −B ( f p ));
where
B
(
f
p
)
=
P
(
f
p
|
y
=
1
)
P
(
y
=
1
)
P
(
f
p
|
y
=
1
)
P
(
y
=
1
)
+
P
(
f
p
|
y
=
0
)
P
(
y
=
0
)
,
f
p
is color intensity of the pixel, P(f p |y=1) is probability of the color intensity equaling f p when the pixel is abnormally-colored, P(y=1) is probability of the pixel being abnormally-colored, P(f p |y=0) is probability of the color intensity equaling f p when the pixel is not abnormally-colored, P(y=0) is probability of the pixel not being abnormally-colored, Y old is Y-component of the color, Y new is Y-component of the dark color, Cb old is Cb-component of the color, Cb new is Cb-component of the dark color, Cr new -component of the color, and Cr old is Cr-component of the dark color.
7. The method of claim 6 , further comprising:
the image processor rejecting at least one non-eye region of the plurality of candidate eye regions.
8. The method of claim 6 , further comprising:
the image processor removing a similar eye region of the at least one verified eye region.
9. The method of claim 6 , wherein the image processor detecting the abnormally-colored regions in the image data comprises:
the image processor utilizing a neural network to predict abnormal pixels;
the image processor grouping the abnormal pixels into a plurality of candidate abnormal regions;
the image processor calculating color intensity of each abnormal color in the image;
the image processor selecting abnormally-colored pixels according to an intensity threshold; and
the image processor forming abnormally-colored regions out of relocated pixels of the abnormally-colored pixels.
10. The method of claim 9 , further comprising rejecting non-pupil candidate abnormal regions of the plurality of candidate abnormal regions.
11. The method of claim 9 , wherein the image processor selecting abnormally-colored pixels according to an intensity threshold is the image processor selecting abnormally-colored pixels having red, blue, or green color intensity greater than 0.5.