Purchase this article with an account.
ST Clay, MJ Moseley, AR Fielder, C Dainty; Automated Photomontaging of Retinal Images . Invest. Ophthalmol. Vis. Sci. 2002;43(13):4343.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To develop a wholly automatic system for registration of images obtained from a Retcam 120 digital fundus camera (Massie Research Laboratories, Dublin, CA) to create a wide-field montage. Single wide field retinal images should aid image interpretation as the examiner would only have to examine one,coherently presented, image. Manual montaging is a time consuming process, and previous attempts at automatic systems[Mahurkar et al, Invest. Ophthalmol. Visual Sci. 1996;37:1675-1883] require significant user input. Methods: Images are standardized by applying image processing techniques (median and homomorphic filtering) to remove noise and to eliminate illumination non-uniformity. Two methods of locating "landmarks" in the image are compared. One method is based on statistical considerations, and one is based on matched filters. Graph matching techniques are discussed for pairing landmarks in separate images. A method of mapping images to the same projection, based on the initial assumption that the eye is spherical, is described. The mismatch of the initial matching will be used to refine this assumption. Results: Comparison between the landmark detection methods suggests that the results of the matched filter are easier to process, but those of the statistical method are more generally applicable. The matched filter method returns structural information regarding the layout of blood vessels, not just locations. The statistical method does not require knowledge of structures present in the image, but does not return as much information. Conclusion: The components of an automated montaging system, a landmark detector, a landmark matcher, and an image mapper have all been described. The components can be integrated into a whole system capable of montaging Retcam 120 digital fundus images. It is envisaged that the techniques described will be extended for use with other retinal imaging systems.
This PDF is available to Subscribers Only