Abstract
This paper proposes an effective noise reduction approach for retinal tomography. Tomographic images include many speckle and impulse noises. The conventional approach removes them by averaging ten or more continuous shots in a second. However, the averaging often causes a blur or disappearance of the diseased part because patient's eyeball moves in three dimensions during the continuous shooting. This paper proposes a new approach based on locally weighted averaging to reduce noise while being diagnosed. Results of the comparative experiments have shown that the proposed approach can hold the feature of the diagnosis part, while the conventional approach loses it.
