Audio Transcript Auto-generated
- 00:00 - 00:03
Hi and welcome to our presentation for signals and systems.
- 00:04 - 00:06
My name is carl and with an and tom our
- 00:07 - 00:08
project was to clean old comic photos.
- 00:10 - 00:12
Comics are an important part of many people's childhoods.
- 00:14 - 00:18
However old comics may have unwanted noise due to wear
- 00:18 - 00:21
or poor print quality Such as DC three images.
- 00:24 - 00:28
So our motivation for doing this was to review GIN
- 00:28 - 00:30
eight old comics by making them more appealing.
- 00:31 - 00:33
We do this by reducing noise in them to make
- 00:33 - 00:34
them higher quality.
- 00:35 - 00:37
Moreover, we want this process to be quick and efficient
- 00:38 - 00:40
so that I can be done for hundreds of comics
- 00:41 - 00:45
as alternative options such as photoshopping or image correcting can
- 00:45 - 00:46
take a lot of time.
- 00:47 - 00:49
Our vision is to encourage people to read and remember
- 00:50 - 00:53
old comics to ensure that remain part of our history.
- 00:55 - 00:57
So the question is how do we create a clean
- 00:58 - 00:59
photo via mala?
- 01:00 - 01:03
Well, we can use the fast fourier transform, we're using
- 01:03 - 01:06
this, we're able to calculate the amplitude from which we
- 01:06 - 01:10
are able to a plot the amplitudes and this allows
- 01:10 - 01:13
us to spot noisy areas and remove them, after which
- 01:14 - 01:17
we're able to reverse for your transform to get improved
- 01:17 - 01:21
image. However, this process only allows one colour of time.
- 01:21 - 01:23
Therefore, at the end we'll add the colours back together
- 01:25 - 01:28
using the fast fourier transform and then calculating amplitudes were
- 01:28 - 01:29
able to get the plot on the right.
- 01:31 - 01:33
This plot shows the noise of the image in the
- 01:33 - 01:36
middle section showed by the curly bracket is crucial information
- 01:36 - 01:38
for the photo without it.
- 01:38 - 01:39
It becomes extremely blurry.
- 01:40 - 01:43
However, the bright areas shown by the arrows are unwanted
- 01:44 - 01:45
noise that we wish to remove.
- 01:47 - 01:49
Initially we thought this was possible to do mainly when
- 01:49 - 01:52
looking at images such as this one we saw that
- 01:52 - 01:55
it was not as these areas are too difficult to
- 01:56 - 01:58
determine and way too many to be able to do
- 01:58 - 01:58
it properly.
- 01:59 - 02:01
Therefore we decided to automate this by about that.
- 02:03 - 02:06
Initially we struggled a lot but finally came to a
- 02:06 - 02:07
following solution.
- 02:08 - 02:10
We wanted to remove all the right areas.
- 02:10 - 02:12
However, protect the crucial one in the middle.
- 02:13 - 02:16
So we created a bound area which is manually inputted
- 02:18 - 02:20
and then this information is stored an empty array.
- 02:21 - 02:23
We then find the amplitudes.
- 02:24 - 02:25
They find the mean of the aptitudes and use a
- 02:25 - 02:29
percentage to find amount of rose to remove the percentages.
- 02:30 - 02:33
Also manual input after which we are able to remove
- 02:34 - 02:34
all the noisy rose.
- 02:36 - 02:38
At the end we get the crucial values again back
- 02:39 - 02:42
to the to the um array from which we removed
- 02:42 - 02:45
the noisy rose from and the impact can be seen
- 02:46 - 02:48
here. As you can see all the black rose should
- 02:49 - 02:51
be deleted once and the middle remains unscathed.
- 02:53 - 02:57
Using the reverse for you fast fourier transform were able
- 02:57 - 02:58
to get the original clean image.
- 03:01 - 03:04
Now we must do this for each color for this.
- 03:04 - 03:07
We must simply repeat the same process three times and
- 03:07 - 03:10
after After the after.
- 03:10 - 03:13
After the inverse for fast food transform, add them together.
- 03:16 - 03:22
This is a result with the initial photos as you
- 03:22 - 03:22
can see.
- 03:24 - 03:27
These are quite much better than the original ones that
- 03:28 - 03:28
I should at the start.
- 03:29 - 03:32
However, this is not good enough for us because we
- 03:32 - 03:32
still want to improve.
- 03:34 - 03:36
You want to automate the percentage of roads removed and
- 03:36 - 03:40
ultimate bound numbers for the crucial values such that you
- 03:40 - 03:42
can easily put any image you want and get an
- 03:42 - 03:46
output without any manual uh, input.
- 03:47 - 03:50
We believe this will improve the quality of photos while
- 03:50 - 03:54
remaining while maintaining the sharpness and allow future generations to
- 03:54 - 03:55
enjoy old comics.