►
From YouTube: An image is worth 1000px - Pierre Roux
Description
"Photoshop" apps allow us to add filters to pictures: effects like blur, sharpen, edge detection, colour segmentation, etc. are all options in these apps, but how does your computer understand how to do this?
Using my interactive slides, I will explain how your computer derives meaning from a grid of numbers (called pixels), how it detects simple features in the image, and then manipulates pixels according to what the user demands. Pay close attention and you will be able to create your own image filters 😉
This talk was recorded at the Jozi.JS meetup group on 24 July 2019.
https://www.meetup.com/Jozi-JS/events/263022711/
Disclaimer: The opinions in this talk are those of the individual and do not necessarily represent those of this channel, the meetup group, or its sponsors.
A
Welcome
everyone,
so
everyone
here
has
used
image
processing
at
some
point
in
their
life.
Maybe
you
added
awesome,
Instagram
filter
in
one
of
your
photos
or
you
modified
one
of
your
photos
for
photoshopped
it
with
one
of
these
expensive
apps
that
you
wouldn't
buy
it
or
you
just
took
a
digital
photo,
but
the
thing
about
it
is:
how
is
it
even
possible,
if
you
think
about
it?
Digital
photo,
it's
binary
numbers
on
a
hard
drive,
but
yet
the
computer,
somehow
figures
out.
A
This
is
an
edge
I'm
going
to
blur
this
H
out
or
I'm
gonna
enhance
this
age
and
show
you
an
outline
of
the
object
in
the
photo.
How
does
the
computer
understand
the
visual
world
today
ultra
ensure
that
using
these
slides
will
learn
how
to
do
image
processing
by
hand
so
to
say
because
pheno
the
concept
you'll
be
able
to
code
it
in
a
system
yourself?
A
So
the
first
question
is:
what
is
an
image
so
remember
back
in
primary
school
we
used
to
have
these
coloring
books.
They
had
outlines
around
regions
and
numbers
in
the
middle.
He
would
color
in
the
number
number
two
is
brown
number.
Three.
Is
green
and
you
would
fall
in
everything
like
that
now
the
thing
about
images
is
they're
really
exactly
the
same
thing,
although
a
lot
more
boring
those
regions,
special
shapes
those
outlines,
they're,
really
just
a
grid,
an
exact
square
grid,
the
numbers
don't
really
even
show
color,
they
show
brightness.
A
A
So
this
will
make
it
much
easier
to
work
with
since
it's
smaller
numbers.
So
let's
see
what
we
can
do.
First,
on
the
left
side,
here
we
have
the
source
image
short
image.
That's
the
image
that
you
want
to
change
in
the
middle.
You
have
the
algorithm,
the
algorithm
you'll
use
to
change
each
pixel,
that's
the
kind
of
thing
that
we
want
to
apply
and
on
the
right
side,
we
have
a
blank
canvas.
This
is
where
we'll
be
writing
out
that
writing
out
our
results.
A
Now
you
don't
want
to
dirty
up
your
source
image
by
writing.
The
results
are
right
next
to
what
you're
working
on
so
write
it
on
a
new
blank
canvas.
What
we'll
be
doing
is
converting
that
source
image
into
black
and
white
a
refresher.
It's
going
to
look
something
about
like
this.
That's
what
we're
going
for
now.
A
The
rules
are
pretty
simple:
we're
going
to
look
at
one
specific,
pixel,
we're
going
to
compare
it
to
the
threshold
value
over
here
now,
if
it's
bigger
than
the
threshold
value,
it
becomes
a
white
pixel,
full
white
7,
the
brightest
that
my
monitor
can
display.
If
it's
less,
it
becomes
a
zero
black
pixel.
The
dark
is
that
it
can't
display.
We,
let
me
just
do
this
on
each
and
every
pixel,
so
on
the
first
pixel
here,
the
value
one
is
one
bigger
than
four.
A
Of
course
not,
so
we
write
down
a
zero
for
the
next
one
to
two
is
not
bigger
than
four,
so
it's
a
zero
three
same
thing:
six
it
becomes
in
string.
Six
is
greater
than
four,
so
we
write
down
a
seven,
and
so
we
continue
through
all
the
pixels
that
might
take
a
while
and
we'll
end
up
with
a
black
and
white
image.
On
the
other
side,
this
is
the
image
that's
been
threshold
at
any
pixel
value.
That's
greater
than
this
threshold
has
been
has
been
highlighted
here
and
everything
else
has
been
darkened
out.
A
Now
you
can
see
there's
a
missing
corner
on
a
tee
day.
That's
because
the
threshold
value
I
really
chose
it
at
ran
vary
any
meenie,
miney
moe.
Now
we
can
change
that.
We
can
tweak
the
threshold
value
if
we
make
it
higher.
It
seems
things
are
getting
worse,
there's
less
pixels
that
are
greater
than
this
bigger
threshold.
Now
the
thresholds
5,
so
obviously
some
pixel
won't
be
bigger
than
five.
A
So,
let's
try
and
tweak
it
lower
over
here
at
three
we
seem
to
be
getting
a
good
result,
so
just
the
right
amount
of
pixels
breach
the
threshold
of
three
to
share
the
image.
If
we
keep
tweaking
it
lower
and
lower
we'll
get
something
worse
now
the
cool
thing
about
this
is,
you
can
add
any
algorithm
that
you
like
you,
can
imagine
up
anything
and
get
some
result
out
of
it.
Imagine
we
want
to
brighten
up
the
image
every
prick,
pixel
value,
plus
one,
do
it
for
all
the
pixels.
A
Now
it's
brighter
minus
one
to
make
a
darker,
it's
kind
of
boring.
You
can
invert
the
image
you
take
the
maximum
color
def,
which
is
seven
minus
the
pixel
value.
So
we
can
look
at
one
of
these
pixels
one.
So
we
say
seven
minus
one
write
down
the
value
over.
They
will
get
an
inverted
image.
It
will
be.
The
colors
will
be
the
other
way
around.
Now
this
is
cool,
but
how
can
we
make
it
better?
What
if
we
want
to
do
something
that
requires
more
pixels
around
it?
A
A
We
could
attack
the
source
pixel,
take
its
eight
neighbors
around
it
and
then
sum
all
those
values
and
divide
them
by
count
sounds
a
lot
like
high
school
find
the
average
a
few
data
points
divided
by
the
amount
of
data
points.
So
what
does
this
look
like?
If
we
look
at
this
one
over
here,
we
see
we
see
the
center
pixel
zero
and,
on
the
other
side,
the
eight
neighbors
a
bunch
of
other
zeros.
A
Now,
if
we
look
at
the
sum
down
there,
it's
a
bunch
of
zeros
plus
1/9,
it's
close
enough
to
one
just
round
it
to
one,
but
that's
again,
because
my
display
can
only
show
eight
different
colored
lives,
so
I
can't
show
decimals.
Now
you
write
down
that
one.
On
the
other
side
you
go
over
to
the
next
pixel.
You
look
at
that
pixel
plus
it's
eight
neighbors.
It's
nine
pixels
in
title
get
this
sum
divided
by
nine
write
it
down.
As
we
move
here.
Stuff
starts
getting
different.
A
We
get
a
value
of
three
on
the
other
side.
This
is
because
there's
much
more
savings
in
the
sum,
so
the
sum
gets
much
bigger
as
we
continue
on
the
values
change
over.
Here
we
get
a
five
now
to
make
sense
of
this.
We
will
need
to
see
the
bigger
picture,
but
before
that,
I
want
to
show
you
something
cool.
If
you
look
at
the
value
for
the
seven
here,
it's
obvious
the
average
is
going
to
be
seven
on
the
other
side,
so
those
are
unchanged.
A
So
what
does
blurr
exactly
do
if
you
look
at
the
zeros
here,
they're
also
unchanged,
but
in
between
on
this
age,
we
get
a
value
in
between
those
two
values.
We
get
a
smoother
value.
What
I
mean
by
the
speed
of
the
gradient
when
going
over
the
picture?
Imagine
is
a
little
bug
running
from
the
left
of
the
picture
towards
the
right
it
doesn't
have
to
be
about.
It
could
be
a
feature
as
well.
Now
it
will
be
on
the
value
0
and
the
next
moment.
A
Suddenly
it's
on
a
value
7,
it's
a
large
change
in
color.
If
we
look
on
the
other
side
here
we
see
it's
been
smoothed
out.
It
goes
from
zero
to
two
five
and
then
seven,
so
this
bug
has
a
less
steep
incline
to
climb,
as
it
goes
over
the
image.
It's
smoother,
there's
no
high
steps
in
it.
So,
let's
see
the
result
of
this
from
here,
you
can
see
all
the
sharp
edges.
All
those
sudden
changes
in
color
have
been
removed.
Well,
they
have
been
smeared
and
smudged
out,
but
that's
blurting.
A
Now,
if
we
think
about
this,
we
can
add
our
own
algorithms
to
this,
make
something
cooler
figure
out
other
effects,
but
it's
kind
of
hard
working
like
this,
using
the
eight
neighbor
pixels
and
the
center
pixel.
How
do
you
make
something
bigger
out
of
this?
How
do
you
generalize
this?
What
if
we
multiply
each
of
those
neighbors
by
one
sounds
silly.
It
sounds
ridiculous,
but
him
out
we
call
this.
The
kernel
we'll
be
multiplying
each
of
those
by
one
now.
The
idea
here
is
you'll.
Take
a
pixel
on
the
Left
image.
A
You
will
take
one
of
its
neighbors,
multiply
it
by
the
corresponding
kernel
and
then
use
the
sum
of
that
divided
by
the
count,
and
then
you
get
an
output.
So
something
like
this
for
this
pixel
we're
going
to
multiply
it
by
one
on
the
other
side,
plus
the
other
pixel
multiplied
by
one
plus
the
other
pixel
multiplied
by
one.
Just
follow
all
the
factors
through
here
you
can
end
up
with
27,
divided
by
nine.
That's
really
stupid.
It's
kind
of
the
same
thing
that
we
had,
but
now
we
can
make
it
better.
A
Now
we
can
add
stuff
to
it.
So
what
if
we
increase
the
white
in
the
middle
here,
that
means
the
middle
pixel
will
have
a
larger
influence
on
the
final
result.
Everything
will
be
multiplied
by
one,
except
the
middle
pixel
gets
multiplied
by
eight
okay,
so
the
normal
result
that
we
look
at
is
this
very
smart.
That's
what
we
used
before
adding
the
middle
pixel
higher
will
give
us
a
result
like
this.
You
can
actually
see
the
shape
inside
there,
but
yet
it's
more
blurred
than
this
one.
A
So
we
have
achieved
blurred
without
destroying
the
image.
That's
a
good
step
now.
One
thing
you
need
to
keep
in
mind
is
when
adding
these
together,
it
divides
by
16,
that's
16!
It's
a
relative
to
the
amount
of
total
weight
inside
here.
So,
if
you
add
all
these
weights
together,
you'll
get
the
number
you
need
to
divide
by
the
total
weight
of
the
kernel.
A
Now
we
can
use
this
to
do
some
awesome
stuff.
A
popular
blurring
filter
is
called
a
Gaussian
filter,
so
this
is
weight
it,
but
it's
weighted
more
smoothly
than
I
did.
It
follows
a
nice
circular,
smooth
bump
in
the
middle.
Basically,
anyway,
at
the
end
of
the
day,
we
wanted
to
do
something
really
cool.
What
if
we
change
this
value
in
the
middle
to
be
negative?
A
What
kind
of
effect
are
we
going
to
get
if
it's
negative
we're
going
to
be
multiplying
these
by
one,
adding
together,
multiplying
by
one
adding
adding,
except
when
we
get
there?
Now
we
get
three
x
minus
four,
so
it
starts
giving
a
weird
effect
in
our
image.
It
starts
showing
the
outline
of
the
region
that
we're
looking
at
this
is
called
a
sharpening
filter
so
for
the
sharpening
filter,
the
color
on
the
image
here,
why
I'm
not
using
greyscale?
A
If
you
take
a
look
at
the
values
inside
the
result,
a
it
ranges
all
the
way
from
minus
nineteen
to
positive
30
I,
don't
have
the
amount
of
colored
F
to
show
that
that's.
Why
I'm
using
extra
color
to
kind
of
show
that,
if
you
think
about
the
change
here
between
zero
to
seven,
how
steep
that
gradient
was
over
there
minus
nineteen
to
thirty
is
like
really
really
super
steep.
The
reason
behind
this
is
this
is
a
highlight
of
the
ages
in
your
image.
It
highlights
where
the
image
suddenly
changes.
Color.
A
We
can
use
this
for
sharpening.
This
is
exactly
what
sharpening
is.
What
are
we
gonna
do
is
add
this
together
on
the
left
side,
we
have
source
image
in
the
middle.
We
have
the
sharpening
guide,
the
one
we
just
built.
It
uses
the
same
kernel
that
we
had
over
there
on
the
right
side.
We
have
the
result
image
now,
adding
it
together
is
really
simple.
We
look
at
one
pixel
over
here.
We
add
it
with
a
pixel
on
the
other
side
and
we
write
down
the
result.
A
A
No,
the
image
on
that
side
can
only
display
up
to
7
from
0
to
7,
so
the
pixel
values
just
cap
or
clip
out
at
those
boundary
edges.
Now.
That's
also
the
reason
why
that
image
looks
are
ugly
on
that
side.
It's
really
hard
to
make
out.
It's
like
over
sharp
and
it's
kind
of
a
deep-fried
meme
at
this
point,
but
we
can
tweet
that
we
can
scale
down
the
sharpening
guide,
give
it
some
lower
values
by
this
I'm,
just
multiplying
each
of
the
values
by
0.3
or
something
around
there.
A
A
Now
I
forgot
to
show
you
guys
what
sharpening
looks
like,
but
you
get
the
idea
now
another
cool
filter
we
can
do
going
off
from
this,
taking
the
kernel
and
having
your
position
working
out
using
the
position
of
a
pixel
relative
to
another
pixel
to
find
its
value.
Instead
of
that,
we
can
use
ordered
statistics.
This
is
where
you
take
the
9
pixels.
He
ordered
them
by
value
in
a
linear
list,
and
then
you
choose
the
middle
value.
The
median
so
we'll
take
these
9
pixels
order
them
by
value.
A
It's
a
bunch
of
zeros
6
7!
Don't
care
about
that!
We
take
the
middle
one
out
of
9,
so
we
take
1,
2
3,
4
5!
Stop
we
take
the
5th
one
doesn't
matter.
What
else
day
is
and
writing
that
down
gives
us
the
result
over
there
for
the
next
one?
We
take
the
middle
of
those
9
pixels
same
thing
over
and
over
again
now
the
idea
behind
this
is
we
want
to
remove
noise
from
the
image
on
the
left
side
image
you
can
see.
There's
some
noise
called
salt-and-pepper
noise.
A
It's
like
really
dark
and
really
bright.
Pixels,
it's
white
and
black
Peaks.
That's
why
the
salt
and
pepper
so
in
really
bad
cameras.
If
you
have
noise
like
that,
if
you
just
apply
a
blurred
filter,
it
will
remove
that
noise,
but
it
will
also
blur
out
the
edges
in
the
image.
This
has
a
way
of
removing
the
noise
without
disturbing
the
regions
that
much
you
still
maintain
your
regions.
Your
object
outlines
in
original
image.
So
on
the
left
side,
there's
lots
of
noise
with
speckles
down
there
and
up
in
the
upper
right
corner.
A
If
we
look
on
the
right
side,
we
can
still
see
the
circle
on
the
left
and
the
rectangle
on
the
right
has
seen
better
days,
but
it's
mostly
still
there,
but
the
greatest
thing
is:
we
have
no
noise
over
here.
It's
completely
clear
now
in
real
life.
Well,
on
a
high
resolution
picture,
it
will
look
something
like
this.
It
looks
like
it
cos,
unifies
the
image,
so
it
mashes
up
all
the
colors
because
it
chooses
the
median.
The
median
is
different
from
the
average
in
the
sense
that,
if
you
add
very
high
values.
A
If
you
add
very
high
values
to
your
dataset,
like
we
have
over
day
with
the
five
and
seven
everything
is
the
era
you'd
expect
that
average
is
close
to
zero.
If
you
take
a
normal
average,
you
would
get
the
sum
of
these
divided
by
nine,
which
won't
be
zero.
If
you
take
the
median
these
outliers,
they
don't
matter,
it
doesn't
change
the
image
it
it
takes.
Some
people
believe
median
is
a
better
average
than
the
actual
average.
Just
because
it's
more
robust
it
doesn't
care
about
outliers.
A
You
get
a
more
true
value
in
all
circumstances,
where
ever
each
easily
gets
thrown
off
by
outliers.
Now,
there's
another
method,
I
want
to
show
you
guys.
This
is
going
to
be
quite
hard
to
believe
it's
called
Fourier
transform.
The
thing
about
this
is
the
astronaut
picture
on
the
left.
Side
can
be
transformed
into
the
store
picture
on
the
right
side.
It
is
the
same
image
with
the
same
information.
You
can
reverse
transform
the
store
image
back
into
an
astronaut
image.
A
Anyone
not
believe
it.
That's
good
skepticism.
Now
the
idea
behind
this
the
easiest
way
to
try
and
explain
it
is
to
not
explain
it.
It's
really
difficult
what
everyone
is
watched
season.
Three,
stranger
things:
it's
like
it's
like.
Looking
at
the
image
from
the
upside
down.
It's
like
looking
at
an
image
from
another
dimension.
You
really
see
it
from
inside
out.
Nothing
makes
sense,
but
once
you
understand,
is
it
completely
makes
sense
the
image
the
star
on
the
right
side?
A
It
is
a
display
of
the
images
properties,
how
much
sharpness
and
how
much
blur
it
has
where
the
sharpness
is
and
where
the
blur
is
so
the
middle
of
the
store,
the
core
of
the
store
that
represents
the
blurry
details
of
the
image,
so
blurry
details
in
the
sense
represents
color
blobs,
any
constant
color
on
the
astronaut
image
that
will
be
represented
in
the
middle
of
the
star
around
the
rim
outer
edge
of
the
soar.
You'll
get
the
high
post
information,
the
sharpness
information,
so
anything
describing
outlines.
A
Sudden
changes
in
colors
you'll
see
that
on
the
outside.
Now,
at
the
end
of
the
day,
it
looks
pretty
cool.
But
what
do
you
do
if
it,
except
for
steganography,
you
can
change
properties
on
the
star
and
then
reverse
transform
it
back
into
the
astronaut.
Then
you
get
the
astronaut
image
without
those
properties,
without
some
sharpness
properties
or
blur
properties,
an
example
for
sharpen.
We
remove
the
blurry
parts
from
the
inside.
We
just
blot
them
out
of
a
black
circle,
and
then
we
reverse
transform
into
the
astronaut
they
may
get
a
bunch
of
outlines.
A
It's
literally
just
outlines
is
now
solid.
Colors
for
Blair
kind
of
the
same
thing
happens.
We
blot
out
everything
except
the
blurry
detail.
If
we
reverse
transform
that
we
get
no
sharpness
in
our
image,
no
outlines!
That's
why
it's
so
blurry
out
here
and
if
you
do
something
weird
you're
gonna
get
a
way.
The
result.
Don't
do
this,
your
family
vacation
photos.
You
never
know
what
you're
going
to
end
up
with
now.
A
A
A
So,
every
day
we're
looking
at
this
row,
I've
converted
this
row
into
a
graph,
a
histogram
function
on
the
top.
Now
this
histogram
function
shows
the
first
pixel
on
the
Left
is
3.
That's
close
enough!
The
second
one!
It's
a
6,
that's
closed
due
to
3
6.
It
seems
fine.
It
seems
like
the
same
row
of
pixels.
Now
what
we're
going
to
do.
A
Imagine
drawing
a
train
Grove
on
the
top
of
those
bars
on
the
top
3
histogram,
connecting
the
peaks
of
every
bar
to
the
next
one
draw
a
smooth
line
over
that
a
smooth
curve
to
fit
over
that
Grove.
Now,
what
we're
saying
is
that
we
can
add
a
bunch
of
sine
waves
to
approximate
that
trend
line.
So
how
do
we
add
sine
waves?
Some
of
you
might
know
this.
If
you
don't?
A
Here's
a
1,
Hertz
wave
and
we're
going
to
be
adding
it
to
a
two
heads
wave,
just
pick
a
specific
point
in
time
like
right
over
there,
you
can
see
the
pink
wave
1
Hertz
is
at
minus
something,
probably
minus
0.3,
plus
the
yellow
one
is
at
zero
at
that
point,
so
you
can
see.
The
result
at
the
bottom
is
also
negative.
It's
also
minus
zero
point,
something
if
you
move
it
on.
You
can
see
here.
The
pink
has
a
large
negative.
The
yellow
is
a
big
positive,
add
them
together.
A
You
get
basically
zero
now
that
we
know
this
yeah
now
over
here
we're
going
to
be
adding
these
waves
together.
These
are
the
waves
necessary
do
approximate
that
way,
that
line
that
train
line,
let
you
imagine,
but
that's
only
half
the
ways.
We
also
need
to
add
these
together.
So
add
all
eight,
these
waves
together
and
then
you're
going
to
get
something
that
follows
that
line
update.
A
Now
it
does
jump
up
a
bit
at
certain
point,
but
if
you
sample
that
wave
at
the
center
of
these
bars
because
remember
a
pixel
doesn't
really
have
a
width,
we
give
it
a
width.
We
say
the
image:
is
this
many
it's
as
big
as
the
screen
is
wide,
so
it's
got
physical
proportions
to
it
that
the
data
pixel
doesn't
have
the
data.
Pixel
only
has
a
position
next
to
it
now,
but
it's
all
it
cares
about.
Now
we
can
look
at
some
other
examples.
A
It
follows
those
and
those
are
a
different
way
if
you
need
to
eat
each
of
these,
so
what
we
did
in
the
Fourier
thingy
previously
removing
some
of
the
sharpness
or
removing
some
of
the
blur.
The
left
side
is
the
blurry
waves.
The
blur
detail
low,
pass
information
on
the
right
side,
the
hypotenuse
sharpness,
so
you
can
literally
remove
one
of
these
waves
and
you
won't
have
that
waves
information
up
there.
So
if
you
remove
things
as
correctly,
if
you
remove
a
very
high
frequency
wave,
you
already
move
very
high
frequency
or
very
sharp
noise.
A
So
any
sudden
change
over
here.
Any
sudden
change
like
that,
a
sudden
change
of
color.
It
will
remove
that
if
we
rip
out
that
wave
and
add
them
all
together
again,
so
it
does
in
a
previous
example.
One
problem
with
this
is
this:
is
one
dimensional
wave,
so
one
dimensional
the
image
is
two-dimensional.
That's
why
I've
only
been
looking
at
one
row
at
a
time
if
we
look
at
all
rows
at
once
to
convert
this
to
a
Fourier
store.
You
end
up
with
two
dimensional
waves.
A
It
looks
like
a
bedsheet
with
waves
in
it,
and
every
pixel
here
in
the
image
will
represent
well
the
store
image
of
the
source
image
every
pixel.
They
will
represent
one
of
these
bed
sheet
waves
in
a
different
direction,
a
different
frequency
and
a
different
phase
angle.
But
imagine
like
a
ripples
in
the
ocean.
If
there's
thousands
of
these
ripples
right
next
to
each
other
and
I
correctly
placed
he
will
get
peaks
and
troughs
at
the
correct
place
so
that
you
can
make
an
image
on
water,
it's
really
impossible
of
the
technology.
A
A
A
Self-Driving
cars
is
becoming
a
really
good
thing.
Ilan
heights
lidar
he
thinks
image.
Processing
is
the
way
to
go
I'm
working
for
that.
That
sounds
cool
image
processing,
but
there's
lots
of
other
use
cases
as
well,
most
new
smartphones.
They
open
up
using
face
detection.
You
get
object,
recognition
using
Google
lanes.
If
you
don't
know
that's
a
chair,
it
can
really
tell
you
that
it's
important
stuff
to
know
there's
more
useful
stuff
in
life,
though
I
found
a
project
that
some
guys
are
working
on
where
they
want
to
make
an
indoor
GPS
system.
A
Now
GPS
works
indoor.
What's
the
problem?
The
problem
is
when
you're
five
stories
in
the
basement
on
the
ground
and
you're
new
at
your
new
job
there
and
you
need
to
find
a
meeting
room
on
the
other
side
of
the
building.
It's
really
cool
if
your
GPS
could
show
you
the
hallways,
so
this
will
be
able
to
use
your
camera
figure
out
where
you
are
based
on
the
always
in
the
camera,
the
stuff
that
it
can
see
around.
You
matches
that
to
a
real
map.
A
That's
been
taken
off
the
building
and
it
can
tell
you
you're
in
this
hallway
this
three
meter
radius.
You
must
be
around
there
and
then
you
can
just
follow
the
path,
the
same
as
I.
Think
half
of
us
that
using
Google
Maps
on
the
way
here,
some
other
cool
things
is
in
astronomy,
so
I
have
to
break
you
guys,
but
most
of
the
stuff
we
see
in
astronomy
is
fake.
Now
the
earth
is
round.
None
of
that.
A
But
the
thing
that
we
see
is
these
pictures
or
false-color.
That's
because
if
you've
seen
some
astronomy,
videos
documentaries,
they
talk
about
telescopes.
Getting
these
images
through
massive
telescopes
and
I'll
show
pictures
to
these
telescopes
and
it's
a
radar
dish
how's
that
a
telescope,
because
it's
looking
at
radio
waves,
it's
picking
up
radio
waves
stuff
that
we
can't
see
it
stole
photons,
it's
electromagnetic
spectrum.
Now,
if
you
color
it
in
that's
radio
waves,
you'll
get
a
picture
just
like
the
pictures.
I've
shown
you,
they
have
a
different
intensity.
A
We
just
need
to
give
the
correct
color
to
the
different
energies
that
we
actually
pick
up
with
this
telescope.
Then
we
can
make
images.
The
idea
of
image
processing
is,
if
you
do
that
correctly,
you
mess
up
your
data.
You
mess
up
your
image.
You're
gonna
make
wrong
assumptions
about
what
you
actually
see
out
there
using
image
processing
correctly.
You
can
color
in
their
stuff
correctly,
so
it
actually
makes
sense
and
you
get
the
truth
out
there.
So
a
lot
of
you
might
remember
the
black
hole
image.
A
The
reason
it's
that
blurry
is
because
there's
no
more
information
to
it.
Adding
trying
to
make
it
sharper
at
this
point
is
adding
information
to
it.
You're
gonna
get
a
false
or
false
truth.
You
won't
get
the
real
truth
there.
It's
supposed
to
be
and
lots
of
the
other
examples
you
can
see.
The
colors
are
quite
bright
and
that
to
make
stuff
obvious
now
the
last
cool
thing
I
found
was
this.
A
Some
guys
in
university
had
basically
tried
to
see
Wi-Fi
I
wanted
to
make
Wi-Fi
vision,
so
this
little
robot
running
on
an
Arduino
with
some
stepper
motors
and
it
picked
up
Wi-Fi
more
or
less
in
a
specific
location,
a
specific
direction,
so
it
with
point
360
degrees
around
it
and
also
said
90
degrees
up
and
down
and
map
an
entire
hemisphere
around
it.
They
basically
made
an
image
that
looks
like
this
at
the
bottom.
Here
is
the
Wi-Fi
image
at
the
top.
A
Is
a
panorama
photo
of
the
same
field
of
view
that
this
Recorder
telescope
thing
he
had
so
they
predicted
to
say
DC
Wi-Fi
waves
around
they,
because
they
Rooter
is
on
the
other
side
of
that
wall.
They
know
this
now
using
image
processing.
You
can
overlay
this
and
get
something
about
like
this.
It
shows
you
the
strength
of
Wi-Fi
around
you
with
a
hotspot
right
there,
another
Wi-Fi.
What
about
a
real
hot
spot?
It's
a
heat
map
of
stuff
around
you!
So
just
imagine
if
they
ever
wins.
The
next
Terminator
movie
coming
out
so
I.