Joe Padfield  0:05  
Maria, if you'd like to continue. 

Maria Villafane  0:07  
Yes, thank you. Thanks. Hello, my name is Maria Villafane, I'm currently a PhD student in conjunction with the National Gallery and Imperial College. I'm living with the multi-modal registration of old masters paintings. (Next please). So this an introduction. I'm working with my supervisors Catherine Higgitt at the National Gallery, and Pier Luigi Dragotti at Imperial College looking at registering, these technical images, as Nathan was describing. Specifically I'm looking at aligning the XRF data cubes to the visible image of the painting. So, (next please). So it's important to say that the way that I'm looking at this problem at the moment is to understand the data cubes. Specifically XRF as a selection of elemental maps sources across entities that are aligned by construction. So, I look at each image at a time. (Next). And, and I basically look for what is the this transformation that I do apply, registered to the visable image of the painting. And I'm looking at Transformations because, it's important to say, that each of these images are depict each elemental amp. Throughout this of this data cube, you will have different features that are not always visible in the visible image of the painting. For example, in this one we can see, for example, that the nose of the dog will be, will be apparant in both the visible image, and one of the slices of this data cube of taking of technical images. (Next), But in most cases, this is a difficult assumption to make. So working with feature detection descriptors might not work. (Next). So myself looking, at a particular features in the visible image and the technical images, I'm actually more than I used to matrix escaped composites, they use it commonly in medical imaging sector, typically looking at which information was submitted for alignment (next) And these will be basically a reference to understand how we'll align our given images for example the blue and the red profiles of this human skull and brain will be best aligned. When we can see that they are separate overlap together and they show in purple, as the combination of red and blue. And, and this will return, the highest metric. So for this case I'm using for implementing the techniques. I'm using a library called Simple Elastix studies, quite well implemented in the medical imaging sector. (Next please). So as I was saying, the X-ray data cube or XRF, I basically use it as a stack of separate images so I use a series of slices of this data cube, and (next) one 'slice', quote unquote, at the time, I search for this transformation for each of these slices for each of these images, for each of these elemental maps. (Next). And the way I operate with this is for each slide, I go, I'm trying to find the best transformation. This is not the next slide, sorry. Okay, that's fine but I can I can, yeah, sorry, I was going to make a comment, but that's fine. So basically for each slice. They basically in brief brush strokes, there is an appendix to actually go further into this topic, but I perform three searches at different resolutions so basically first doing a very sparse search, then a, how do you say like, a search that is not that sparse, and then an exhaustive search at the very end. What I look for in each of the searches is the best location. What are the highest metric possible of mutual information. This is what this is all for each slide. And the way to encpass, let's say like a unique transformation for the data cube, can be to take each of the transformations and look for the highest average of the mutual information for the data cube as a whole. So, (next). And yeah, and this is, this is basically in a nutshell the process that I'm working with. Thank you very much.