Title: Segmentation of retina images to detect abnormalities arising from diabetic retinopathy

Authors: Amrita Roy Chowdhury; Sreeparna Banerjee

Addresses: Department of Computer Science and Engineering, Maulana Abul Kalam Azad University of Technology, Salt Lake, West Bengal, India ' Department of Natural Science and IEM, Maulana Abul Kalam Azad University of Technology, Salt Lake, West Bengal, India

Abstract: Segmentation of retina images to isolate and detect abnormalities is an important step before the classification can be performed. In this paper, we apply three popular unsupervised segmentation algorithms, namely, K-means clustering, fuzzy C-means clustering and Otsu multilevel thresholding algorithm to extract dark and bright lesions caused by diabetic retinopathy, in the earlier stages of its prognosis. This segmentation process also helps in removing normal structures in the retina images like optic disc and blood vessel tree. The results of the best performing segmentation algorithm can subsequently be used in classification and thereby aid the ophthalmologists in diagnosing the disease. It is found that while Otsu segmentation performs best, K-means is a close second and outperforms fuzzy C-means clustering in terms of time complexity and is therefore the best choice.

Keywords: fuzzy C-means; FCM; K-means; Otsu algorithms; retina image segmentation.

DOI: 10.1504/IJIM.2021.115745

International Journal of Image Mining, 2021 Vol.4 No.1, pp.36 - 45

Accepted: 22 May 2019
Published online: 21 Jun 2021 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article