May 2007
Volume 48, Issue 13
Free
ARVO Annual Meeting Abstract  |   May 2007
Composition of Wide Field Fundus Images Using a Non-Linear Registration Technique
Author Affiliations & Notes
  • D. Baumgarten
    Institute of Biomedical Engineering and Informatics, TU Ilmenau, Ilmenau, Germany
  • A. Doering
    R & D Software, Carl Zeiss Meditec AG, Jena, Germany
  • M. Trost
    R & D Software, Carl Zeiss Meditec AG, Jena, Germany
  • Footnotes
    Commercial Relationships D. Baumgarten, Carl Zeiss Meditec AG, E; Carl Zeiss Meditec AG, P; A. Doering, Carl Zeiss Meditec AG, E; Carl Zeiss Meditec AG, P; M. Trost, Carl Zeiss Meditec AG, E; Carl Zeiss Meditec AG, P.
  • Footnotes
    Support None.
Investigative Ophthalmology & Visual Science May 2007, Vol.48, 2584. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      D. Baumgarten, A. Doering, M. Trost; Composition of Wide Field Fundus Images Using a Non-Linear Registration Technique. Invest. Ophthalmol. Vis. Sci. 2007;48(13):2584.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract
 
Purpose:
 

The composition of retinal images presents high demands to the applied methods. Substantially different lighting conditions between the images, glarings and fade-outs within one image and nonlinear distortions are the main challenges. We present a fully automatic algorithm for the registration of retinal images and their overlay to montage images combining area- and point-based approaches.

 
Methods:
 

In a first step the spatial relation between all images is approximated. For that purpose the cross correlation is applied to preprocessed images, in which the vessels as the most prominent structures in fundus images were enhanced. Based on this prepositioning, point correspondences between the images are determined by a blockmatching algorithm that applies a diamond search strategy in different resolution levels. A method for choosing suitable candidate points using a double threshold criterion is presented. It ensures a uniform distribution as well as a sufficient quality of these points. The Normalized Correlation Coefficient serves as the similarity measure in both steps as it proved to be more robust and less time consuming to compute compared to Mutual Information which was also examined. Subsequently a quadratic transform model is adapted to the correspondences that accounts for the roughly spherical surface of the retina. It was chosen out of 4 models based on visual and numerical evaluation. Finally the transformed images are overlaid. In doing so the pixels are weighted to smooth the seams between the images.

 
Results:
 

The algorithm was validated using a total of 40 sets of fundus images of healthy persons as well as images with clear pathological disorders. For 36 sets all images could be composed. The computed montage images show no misregistrations. The mean registration error is less than 2 pixels for all sets. Shown below is a montage of seven fundus images computed with our algorithm.

 
Conclusions:
 

Our algorithm allows the composition of retinal images to wide field montage images. It approved to be very robust with respect to the quality and the non-uniform illumination of the images.  

 
Keywords: image processing • retina 
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×