Good afternoon.

Today, in the upcoming 45 minutes we'll continue to

discuss methods for defect pixel interpolation.

I just opened the slides,

which explains to you

which type of images we are currently considering.

We have here an x-ray image, here you see a pacemaker,

you see the chest, you see the spine,

and then you see areas where we have defect pixels.

And these defect pixels, they appear in systems

where we use flat panels.

Flat panel detectors are very hard to be manufactured,

so it's a tough time to get a detector up and running.

And sometimes they have to deal with defects.

And minor defects can be eliminated

by using proper image processing methods.

And that's exactly what we are currently looking into.

How can we design algorithms that take

the information where the defects are,

consider the pixels in the local environment

or global environment, whatever,

and then they try to estimate what is the pixel information

here at the defect using the neighboring pixels.

And this is called defect pixel interpolation,

computing the pixel information for those pixels

where the detector is not doing any measuring

or any measurements.

So very intuitive concept.

And then last week, we started to look

into a mathematical model.

How can we model this process of generating images

with defects?

How can I model this in, or how can this be modeled

in terms of a mathematical model?

And it's pretty easy to characterize.

We have here the ideal image.

That's the image without any defect.

Then we have our defect mask

where the mask says I'm one if the pixel can be measured.

I'm zero at this point here if the pixel cannot be measured

if there is a defect.

What you can do then, you can take this image

and you can multiply this image component-wise

with the ideal image.

And what you get is basically the defect image.

The defect image.

So sounds quite reasonable, right?

Multiplication with a mask image.

And what do we have to do?

If we want to compute the original image