Music

All right.

And yeah, welcome to Multimedia Security.

Today I will walk through a number of methods

that are specific for the JPEG format.

Some image forensics methods that make use of the JPEG format.

And the papers that I've selected,

they are all from the group of Honey Farid.

But there are also other groups,

like the group by Alessandro Piva.

We had a look at a couple of their papers

for the watermarking topic.

They also did a lot on JPEG forensics, for example.

There's also Qi Wu Huang from China.

So looking specifically into the JPEG format

to detect manipulations in images has been a popular topic.

First of all, how does the JPEG compression work?

We already talked about this,

but let's restate this.

So the JPEG encoding

roughly works like this.

We have as input,

we have a three channel RGB image.

And the steps are convert RGB

to another color space that is called YCBCR.

So Y is the luminance, brightness.

And CB is the blue chromaticity.

And CR is the red chromaticity.

The idea is that once we have the luminance

singled out or separated from the colors,

we can apply compression of different strength.

So we are a little bit more careful with the luminance

because the eye is more sensitive to luminance changes.

And we can apply a stronger compression

on the blue and red color chromaticity.

So green is missing because the green chromaticity

is what is missing in order to add these values

up to unity or to 255 depending on what your basis is.

Okay, the second step is to subsample each channel.

That's an optional step.

Subsample each channel

by a factor of two, three, four.

So these are the typical factors.

In principle, you could do more.

So this subsampling is typically done

on the color channel also,

but it's arguing that we don't see these changes anyways

so much.

And then third step is to subdivide the image

into eight by eight blocks