08/4 Mosaicing

		Mosaic of seven fundus images

A key element of the aforementioned computer assisted retinal laser photocoagulation system are the images of a retina used for treatment planning and as a spatial map during surgical treatment. As the fundus images of the human retina are generally weak in perspective, multiple of those images have to be fused to form a single mosaic covering the entire domain of the treatment area.

Building a seamless mosaic from a sequence of weak-perspective images has been studied for over two decades. However, its application for fundus images poses special problems that must be addressed:

• The classical region detectors and region descriptors commonly used are bound to fail, as the self similarity of the found regions is very high and the structural content in those images is generally low.
  
• The images are taken with uncalibrated cameras showing high perspective distortion due to the proximity of the objects.
  
• In contrast to most other mosaicing applications, the resulting mosaic should not only be visually appealing but geometrically distortion free. This is important, as the mosaic is not only used during treatment planning, but also as a spatial map during the surgical intervention.
  

The objectives of this project are thus:

• to find suitable application specific region detectors and region matching techniques
  
• to define an accurate mathematical model describing the image-to-mosaic transformation
  
• to assess multiple blending techniques, such as multiband blending, to guarantee a seamless mosaic.