Abstract
When digitized fundus images are used in medicine, important information for clinicians' diagnosis should be maintained well after scaling. Conventional scaling methods select the interpolation kernel based on Shannon's sampling theorem which is only appropriate for band-limited signals and usually generate image with jaggy noise or blurred blood-vessel shapes that the tortuosity and diameter change of blood vessels which are important information to clinicians' diagnosis can not be maintained clearly. To solve these problems, we select the quadratic Fluency sampling function as the interpolation kernel to maintain the color reality of fundus image based on the Fluency information theory. Scaling is then realized by interpolation directed to the blood-vessel shape map of required resolution which is generated from two kinds of Fluency sampling functions. These two kinds of Fluency sampling functions are precedingly utilized to approximate shapes of blood vessels on the original fundus image according to the suggestion of specialists in fundus field. Resulting images by the proposed method show good shape maintenance of blood vessels and have a better quantitative evaluation than the generally recognized best conventional method.
