Reconstructing geometrically consistent tree structures from noisy images
MICCAI'10 Proceedings of the 13th international conference on Medical image computing and computer-assisted intervention: Part I
| Hi-index | 0.00 |
We present an automatic method to segment the blood vessels in retinal images. Our method is based on tracking the center of the vessels using the Kalman filter. We define a linear model to track the blood vessels, suitable for both the detection of wide and thin vessels in noisy images. The estimation of the next state is computed by using gradient information, histogram of the orientations and the expected structure of a vessel. Seed points are detected by a set of matched filters in different widths and orientations. Tracking is carried out for all detected seed points, however we retrace the segmentation for seeds with small confidence. Our algorithm also handles branching points by proceeding in the previous moving direction when no dominant gradient information is available. The method is tested on the public DRIVE database [10] and shows good results with a low false positive rate.