Exudates are one of the principal lesion present in the normal development of Diabetic Retinopathy (DR), its detection is an important step in (DR) screening and classification. This paper presents an automated method for bright lesions detection in retinal images by means of the mean shift filtering. Due to uneven illumination of retinal images it is necessary to perform a preprocessing step consisting of a shade correction technique finding non-structures pixels and adjusting a third order polynomial to be substracted from the original image. The mean shift filtering is applied to enhance bright areas and to uniform background non-structures regions. A region growing algorithm is performed from local maxima regions taken as seeds to get the final results. A set of 20 retinal images selected and manually tagged by a retinal specialist ophthalmologist were used for the evaluation. Results present a true positive rate (TPR) of 0.627 and a specificity SPC of 0.979. It is demonstrated that Mean shift filtering is a promising method for exudates detection. © 2013 Springer-Verlag.
CITATION STYLE
Cárdenas, J. M., Martinez-Perez, M. E., March, F., & Hevia-Montiel, N. (2013). Mean shift based automatic detection of exudates in retinal images. In Advances in Intelligent Systems and Computing (Vol. 184 AISC, pp. 73–82). Springer Verlag. https://doi.org/10.1007/978-3-642-32384-3_10
Mendeley helps you to discover research relevant for your work.