Blood Vessels and Exudates Segmentation in Eye Fundus Images based on Fourier Filtering

This paper presents a Fourier transform approach to segment blood vessels and exudates in eye fundus color images. The basic idea consists in an illumination enhancement using an homomorphic filter, due to non-uniform illumination conditions in the eye fundus image capture. The design of parametric Butterworth bandpass filters in the Fourier domain is applied to the green channel only, to distinguish the foreground objects from the background. To find blood vessels, the negative of the filtered image is determined to emphasize them, after which a DoG filter is applied to intensify edges belonging to the foreground. The resulting image is binarized using Otsu's method and segmentation is accomplished using a morphological closing operation and a masking operation. On the other hand, to find exudates an auxiliary local statistics image is computed based on the previous Fourier filtered image. The resulting grayscale image enhances the exudates and Otsu's method is again applied to binarize it, and a masking operation gives the corresponding segmentation. Illustrative examples taken from a clinical database are included to demonstrate the capability of the proposed method.