Joe Padfield  0:03  
We're going to begin with Andrew Bruce from the National Gallery, Andrew, would you like to carry on, please.

Andrew Bruce  0:09  
Hello and thank you for having me. If we could go to the next slide please. And just another reminder of who I am. And the next slide please. So yes, so I'm a photographer at the National Gallery, and the kind of registration I work with is predominantly mosaic. So image to image registration to create composite images with higher spatial resolution than if they were captured with just one image. So I capture, process, and create the composite images working with visible light, and X-radiography images, and we produce two to three mosaiced images per week, which gives you an idea of the production level, which I think is quite an important point. Next slide please.

Unknown Speaker  0:59  
We try to capture every painting. I suppose visible, X-ray, and a new process of photo macrographs that captures a painting at 22,638 ppi, with a focus stacking technique, kind of, vastly complicates this, as the capture and registration process is, well you know, the number of images, multiplies vastly. Next slide please. What I really want to champion is that for precisely registered images, precise imaging is essential. And I believe the way that you capture images will always have a direct relationship to the process of registration, and, you know, for example, how even is your illimination is your capture device tracking, or is it panning, this will have a, you know, probably quite a large impact on the way that your images register. Next slide please. So image to image registration for mosaicking is performed using PT GUI, which in many ways is a really great piece of software, and we do get amazing results. But the software is not really designed for this kind of work. It's designed actually for making spherical panoramas. So we kind of wrangle with the software to get it to do what we want and to achieve the accuracy that we need. This makes for a very time consuming and not so repeatable process. The software actually works by finding matching features in the overlap between images, and this is problematic, as errors are cumulative, and changes in magnification, caused by paintings being not perfectly flat  (and they never are), results in geometric distortion in the images. So I've just, I've got some image kind of examples here that just show, for example, areas in a painting of just flat colour, very challenging because there are a few features which grab belongs to, and a little example where, for example, I've got a feature of a Holbein's The Ambassador's painting. This painting has quite a big of warp. And that resulted in some kind of different kinds of geometric distortion in the image, when we created the composite. Next slide please.

Andrew Bruce  3:37  
That's not the right side. That's the one. Yeah, thank you. So, um, unless an x-ray beam can completely cover a painting, and be left stationary, and when you move close to the X-ray beam and the imaging device there will always be some kind of parallax effect. It's not an error. It's just what happens when you create a three dimensional image and represent it, two dimensionally. So while this effect can be minimised through capture, it's likely there will always be some features that do not perfectly register. And I've made this little animation which hopefully shows this quite clearly. So this is one of the challenges we faced when registering X-rays. And my last slide. Next please. Still lastly, I wanted to just kind of throw out an ask: how do we assess and also express the precision with which images are registered. I've included three examples here, one from from some photogrammetry software, one from PT GUI showing actually distancing pixels. And then lastly, like a visual showing where the seams are in images so that you can actually visually assess them. Thank you, that's me done.