Modified Hausdorff-distance-based methods are effective for rapid, accurate montaging of multiple, partially overlapping fundus images. Transformation functions as derived from edge-detected, binarized images allow for construction of a single data set (Figure 3). For photographic and angiographic data where the vessel gray levels vary from light to dark, and intensity-based correlation methods fail, image-preprocessing with smoothing, edge-detection, and thresholding facilitates registration. Following image pre-processing, Matrox inspector software (using a simple template matching algorithm) allows for identification of corresponding regions in photographic and angiographic images. Translation-only transformation functions are thus defined, and the utility of edge-based registration methods is demonstrated.
Non-real-time registration (
CPU seconds) is achieved
by non-optimized Hausdorff-distance-based (translation, rotation, and
scale) algorithms performed on edge-detected fundus photographic and
angiographic images, and on images of a model eye.
Figure 4 depicts the result of superimposing the vessel
skeletons of the angiographic data onto the photographic image,
demonstrating high-fidelity registration of photographic and
angiographic data (corresponding to the images in Figures
1 and 2).
Image overlay is demonstrated by simple image superposition on a computer monitor. In addition, image overlay is accomplished on the operating-microscope-based system described. Photographic images of a model eye were acquired and edge detected. The edge-detected images were then rendered in green, and overlayed on a real-time biomicroscopic image of a model eye (Figure 5). The real-time view through the binocular objectives was then a merged image of the green edges and the real-time model eye fundus image. Accordingly, proof-of-principle for an ophthalmic augmented reality environment is demonstrated.