Most prior art image fusion algorithms are based on two assumptions. First, N-channel input images (e.g. the N=4 channel R, G, B and near-infrared) are fused into a single channel equivalent fused image . And, second, that – other than the fused data capturing all the detail of the channels – there is not a simple structural relationship between the fused image and any of the N-channel input images.
In this PhD project, the successful candidate will develop a guided colour image fusion algorithm that maps N-channel images to M-channels (where M<N and is typically 3 i.e. a colour image). Further, the fused image will be guided by the structure in the original colour image. For the RGB+NIR example, the fused image will ‘look like’ the original image but with some extra detail composited from the NIR channel. The prior art “spectral edge” image fusion algorithm  can be viewed as a guided colour image fusion algorithm but, there, the concept of guide is at a very low level. The project will develop image fusion algorithms based on higher level semantic concepts including image relighting  and semantic decompositions . The image fusion research will take place in the context of the wider ‘Future Colour Imaging’ project .
The successful candidate will join the Colour & Imaging Lab (the C&IL) in the School of Computing Sciences at UEA. The C&IL is one of our largest labs and currently comprises 10 PhD students and 3 postdoctoral fellows.