Image Processing Reference
In-Depth Information
Fig. 6.5 The proposed system
Saliency Map Computation: Fusing Temporal Image Signature approach with
Face Detection Given a video clip with size
m
n
is the image
size and
t
is the number of frames being processed, TIS can be written as follows.
Saliency map as computed in TIS approach [
29
] in Eq. (
6.5
) in the computation
of saliency map (Sect.
6.2.3.1
).
n
t
where
m
SalMap
ð
t
Þ
¼
hMap XY
ð
t
Þþ
vMap XY
ð
t
Þ
Transformation of the detected face to the face conspicuity map is done using
weighted function as proposed in [
5
]. This can be written as
Detected face
!
Transform
fMapXY
t
ð
6
:
6
Þ
Weighted face map can be represented as
wfMapXY
t
¼
fMapXY
t
þ
w
ð
:
Þ
6
7
Saliency map can be combined with face map by adding Eqs. (
6.5
) and (
6.7
)
F
SalMap
ð
t
Þ
¼
S
alMap
ð
t
Þþ
wfMapXY
t
ð
6
:
8
Þ
where hMap and vMap represent horizontal and vertical maps, fMap, SalMap are
the face and saliency maps and
w
is the assigned weight mask (gaussian or linear).
Conclusion and Results It has been shown that the proposed method is a well
appropriate algorithm for not only detecting salient regions but similarly is valuable
tool once combining salient information with human face concurrently in a video
scene. Face definitely attracts human visual attention. Eye tracker results in
Sect.
6.3.1
provide evidence of that fact. The accuracy of the proposed method
depends strongly on the quality and accuracy of face detection method. It is also
