Noise reduction for ellipse fitting on medical images

Proposed is a simple yet effective method to reduce noise for ellipse fitting to scattered data on medical images. The method is developed specifically to address the challenge due to variation in the sharpness, magnitude and continuity of edges, when studying the medical images of a large population. The proposed method exploits prior knowledge to eliminate the inhomogeneity within the region of interest, and uses a k-means clustering technique to eliminate background noise. The objective is to create a `noise-free` edge map for ellipse fitting. The method has been applied to retinal fundus images for optic-disc segmentation. Experiment results demonstrate the benefits of the proposed method.