Abstract
This paper presents a new robust approach for the automatic location of the optic disc in retinal images. We detect several candidates independently of optic disc and the main blood vessels (arcades). Candidates are sorted by reliability. The space of all possible pairs disc-arcades is searched using a priori anatomical knowledge, selecting the pair formed by the most reliable candidates satisfying best anatomical constraints. This pair includes the best optic disc location. The approach was tested using three public available data sets: STARE, DRIVE and DIARETDB1 and 20 SLO images from OPTOS plc, with all data sets containing different pathological lesions of various types and severity. We achieved 91.4% detection rate in the STARE data set (which includes 61.7% of images with severe lesions), 95.0% in the DRIVE data set (15.9% of images with pathologies), 96.7% in the DIARETDB1 (94.4% of images with pathologies) and 95% in the OPTOS data set (75% of images with pathologies).
Keywords:
Acknowledgements
Thanks go to the STARE, DRIVE and DIARETDB1 projects for making their images publicly available. This work was supported by a Northern Research Partnership (Joint Research Institute on Computational Systems) studentship and Optos plc.