Beruflich Dokumente
Kultur Dokumente
Slide 5*
First, we add face anti-spoofing. This is because face recognition its self
can be deceived by a spoof face, whether from printed-photograph,
digital photo from devices, mask based, et cetera. This is the main
problem for face recognition based attendance system.
Slide 6*
Second, we add notification system in order to let the student parents
know the information of student attendance status in certain course.
We choose to notify it through SMS Gateway. The reason is because
the only way to communicate to remote area is through SMS Method.
One of most known and most widely used provider in Indonesia,
already cover almost 95% area of Indonesia.
Slide 7*
But we want to limit the scope of our research. For face-anti spoofing
module, we only focus on developing, implementing, and testing our
face-anti spoofing. We would not go deep into face recognition
explanation.
Also, so far, we only achieve face anti-spoofing to overcome spoofing
attack for digital device media like smartphone or tablet. We havent
done it for another media.
For SMS notification module, we design, implement and test the each
of system module.
Slide 8*
This is the main component of our system. We use Raspberry Pi 3
Model B as a core or a computer that will manages the system like
software and databases. We connect all of the other component there.
Then, a Logitech 270C USB Camera acts as a programmed camera that
will capture the identity of the user through their face.
LM2596 Stepdown Converter, we use it to stabilize the voltage of
SIM800l module.
And the SIM800l, acts as SMS gateway.
Slide 9*
In order to test system, there a scenario for each of component. For
face-anti spoofing, we test it by experiment. For notification system, we
test by black box testing
Slide 10* -> 11*
So this is our system flowchart, as I stated before, we only focus on
face-anti spoofing step here.
Slide 12*
To test Face anti-spoofing module, weve design two experiments. First
observing 10 different face with its spoof face in order to find the
threshold that distinguish between real face and spoof face. We use V
Channel of HSV color distribution only. After we find the right
threshold, next we do experiment II. We test it at different time, which
are in the day and night in order to examine right lighting condition for
our system.
Slide 13*
This is the histogram of v channel distribution of HSV Color for a real
face. It can be seen that most of the value are tend to evenly
distributed
Slide 14*
But for spoof face, most of its value are piled up at bin 255. Lets go
deep into the table.
Slide 15*
We take 10 most v channels value in the histogram. It counts by images
pixel in OpenCV. we clearly can see that theres huge difference here.
The value for real face is evenly distributed at 307 356. But for spoof
face, almost all of the pixels value are piled up at bin 255. So this is our
cue for determining the right threshold to distinguish between real face
and spoof face.
Slide 16*
This is how we determine the threshold, we take that 10 most v
channels value. Then we multiply it by its bin position in histogram,
later we sum it all. There we got the value. Next we need to set right
threshold to limit the minimal detection value for spoof detection. So
we do the experiment I
Slide 17* & Slide 18*
From this line charts we can see that for a real face, sum of 10 its
highest v value ranged between 15000 32000 pixel. Meanwhile for
spoof face, its value ranged between 110000 and 240000 pixels. From
here, we can set the threshold between 32000 110000. For example,
for our project, we set the threshold at 60000, so whenever the value
above 60000, it will be detected as spoof face and vice versa.
Slide 19*
Slide 21*