Multiscale analysis revisited: Detection of drusen and vessel in digital retinal images

In order to more accurately detect drusen and vessel from retinal fundus images, we proposed a set of new features and presented a learning based detection scheme. These features are designed to describe the variation patterns of image's local geometrical structure across various scales. Theoretical analysis and a series of preliminary experimental results demonstrate the extra ability and high accuracy of these features in distinguishing drusen and vessel from the background.