Joe Padfield  0:03  
Okay, if we can hand on now to Nathan daily please Nathan.

Nathan Daly  0:06  
 Thanks so if you can go to the next slide. So I'm Nathan Daly, a research fellow in the scientific department at the National Gallery, and I'll also be representing my colleagues Catherine Higgett and Marta Melchiorre  who are shown here. So building from Andrew's introduction, at the National Gallery we perform a number of types of imaging from visible light photography of different magnifications, which includes photomicroscopy to infrared, and X-radiography, and additionally. (Next). In the scientific department and conservation department, we perform 3d imaging, and (next) also spectroscopic imaging, namely, a macro XRF scanning and reflectance hyperspectral imaging, which generate data cubes that can be thought of as image stacks (next). So we're interested in, sorry, this is not the next slide, I think the slides might be out of order. Go to the next full slide, or you might have skipped it...

Okay, well I can I can talk about difficulties, that's fine. Some of the difficulties that we face, include that the objects we work with, paintings, are actually 3d objects, they're not flat, they have many layers, each of which, or components of which, may contribute to the resulting image depending on the technique with different imaging modalities. So (next). If you'd like to co register them. We have to be aware of different intrinsic distortions, the type of signal and the resolution of each technique. (Next), because most of these edging types cannot be captured cannot capture like paintings and one acquisition, we also have to worry about mosaicing and how that can propagate error as Andrew mentioned, And this may require special treatment of regions overlapping. (Next), finally, most registration approaches currently in use are feature basis, which could be challenged by some of these difficulties that are highlighted. So for example, some images may not have features to align, and because of the different resolutions of each modality, it can be hard to find common features. (Next), and then other difficulties just include the sheer size of the datasets and the associated computational power that would be needed to register them, and how we document these types of data manipulations. And so I'll just hand over next to Maria, who is going to explain one way in which we're trying to address this. So that's done.