TY - GEN
T1 - Automatic Segmentation of Fovea and Macula in Retinal Fundus Images
AU - Mallya, B. Vaibhav
AU - Gagan, J. H.
AU - Chhabra, Garvit
AU - Kamath, Yogish S.
AU - Kuzhuppilly, Neetha I.R.
AU - Kumar, J. R.Harish
N1 - Funding Information:
This work is supported by the Science and Engineering Research Board (SERB)—Teachers Associateship for Research Excellence (TARE) fellowship (Grant No. TAR/2019/000037).
Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - We propose an automated method to segment the fovea and macular region in retinal fundus images which is a precursor to assessing the severity of age-related macular degeneration. The fovea region has been localized using the zero-mean normalized cross-correlation technique. The segmentation of the fovea and the macular region has been achieved using the elliptical active disc technique. The elliptical active disc is a shape-specific active contour model and has five free parameters. To obtain the optimal active disc fit on the region of interest to be segmented, the active disc energy has been optimized with respect to five free parameters using the gradient ascent algorithm. Further computational savings has been achieved using Green's theorem. The experimental validation has been done on MESSIDOR, DIARETDB0, DIARETDB1, DRIVE, and IDRiD fundus image databases adding up to 508 images for segmentation of fovea. We attain an average Dice similarity index and fovea segmentation accuracy of 80.45% and 99.66%, respectively on varied fundus image data.
AB - We propose an automated method to segment the fovea and macular region in retinal fundus images which is a precursor to assessing the severity of age-related macular degeneration. The fovea region has been localized using the zero-mean normalized cross-correlation technique. The segmentation of the fovea and the macular region has been achieved using the elliptical active disc technique. The elliptical active disc is a shape-specific active contour model and has five free parameters. To obtain the optimal active disc fit on the region of interest to be segmented, the active disc energy has been optimized with respect to five free parameters using the gradient ascent algorithm. Further computational savings has been achieved using Green's theorem. The experimental validation has been done on MESSIDOR, DIARETDB0, DIARETDB1, DRIVE, and IDRiD fundus image databases adding up to 508 images for segmentation of fovea. We attain an average Dice similarity index and fovea segmentation accuracy of 80.45% and 99.66%, respectively on varied fundus image data.
UR - http://www.scopus.com/inward/record.url?scp=85145667526&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85145667526&partnerID=8YFLogxK
U2 - 10.1109/TENCON55691.2022.9977929
DO - 10.1109/TENCON55691.2022.9977929
M3 - Conference contribution
AN - SCOPUS:85145667526
T3 - IEEE Region 10 Annual International Conference, Proceedings/TENCON
BT - Proceedings of 2022 IEEE Region 10 International Conference, TENCON 2022
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2022 IEEE Region 10 International Conference, TENCON 2022
Y2 - 1 November 2022 through 4 November 2022
ER -