DRIVE Database
Digital Retinal Images for Vessel Extraction (DRIVE) database consists of 40 retinal images, out of which 33 images are healthy, and the remaining 7 images are affected by certain pathologies. The field of view of the fundus camera that was used to capture these images is 45 degrees and the resolution of images in this database is 565 X 584.
Register on the database providers' website and join the DRIVE challenge to get access to this widely used retinal image database.
Reference: Staal et al. (2004)
STARE Database
STructured Analysis of the Retina (STARE) database was created by scanning and digitizing the retinal image photographs. Hence, the image quality of this database is less than the other public databases.
The images of the STARE database were captured by a narrow field of view of 35 degrees camera and have a resolution of 700 X 605 pixels.
Reference: Hoover et al. (2000)
MESSIDOR Database
Methods to evaluate segmentation and indexing techniques in the field of retinal ophthalmology (MESSIDOR) database consists of1200 retinal fundus images captures using a 3CCD video camera on Topcon TRCNW6 nonmydriatic retinography having a field of view of 45 degrees.
The images of the MESSIDOR database have three different resolutions; 1140 X 960, 2240 X1488, and 2304 X 1536 pixels. Recently, 13 duplicated image pairs were discovered in this database. Also, certain inconsistencies in the grading of images were reported, which the database providers have acknowledged on their website.
References: Abramoff et al. (2013), Decenciere et al. (2014)
MESSIDOR-2 Database
The Messidor-2 dataset is a collection of Diabetic Retinopathy (DR) examinations, each consisting of two macula-centered eye fundus images (one per eye).
Messidor-Original consists of all image pairs from the original Messidor dataset, which is 529 examinations (1058 images, saved in PNG format). Eye fundi were imaged, without pharmacological dilation, using a Topcon TRC NW6 non-mydriatic fundus camera with a 45-degree field of view.
Only macula-centered images were included in the dataset. Messidor-Extension contains 345 examinations (690 images, in JPG format).
Overall, Messidor-2 contains 874 examinations (1748 images). The dataset comes with a spreadsheet containing image pairing.
References: Abramoff et al. (2013), Decenciere et al. (2014)
DIARETDB0 and DIARETDB1 Database
Standard Diabetic Retinopathy Database Calibration level 0 (DIARETDB0) database consists of 130 retinal images taken using a digital fundus camera having a field of view of 50 degrees. Out of 130 images, 20 are healthy and the remaining images contain symptoms of DR.
Standard Diabetic Retinopathy Database Calibration level 1 (DIARETDB1) database has 89 retinal fundus images. Out of 89 images, 84 retinal images contain signs of mild non-proliferative diabetic retinopathy (NPDR), and the remaining 5 are healthy. The field of view of the camera that was used to capture the images is 50 degrees.
The image resolution in both DIARETDB1 and DIARETDB0 database is 1500 X 1152.
References: Kauppi et al. (2007), Kauppi et al. (2007)
REVIEW Database
Retinal Vessel Image set for Estimation of Widths (REVIEW) database was developed by a group of researchers from the University of Lincoln, UK. This database was specifically developed to act as a reference for various vessel segmentation algorithms.
This database has information about the vessel width of 16 images included in this database. The marking of vessel width was done by three independent experts. It contains 193 marked vessel segments in total. The database is divided into four sub-databases: high-resolution image set (HRIS, eight images), vascular disease image set (VDIS, four images), central light reflex image set (CLRIS, two images), and the kick points image set (KPIS, two images).
Reference: Al-Diri et al. (2008)
ROC Database
Retinopathy Online Challenge (ROC) microaneurysms database was developed as a part of an online competition held by three well-known researchers working in the field of retinal image processing. This competition was conducted to identify the best algorithm to correctly identify the microaneurysms in the retinal images.
The database consisted of 100 retinal images, which were divided into two classes; a training class and a testing class containing 50 images each. A reference standard is provided for the training set images indicating the locations of the microaneurysms. The images were captured using Topcon NW 100, a Topcon NW 200, or a Canon CR5-45NM saved in JPEG compression format.
Reference: Niemeijer et al. (2010)
CHASE Database
Child Heart and Health Study in England (CHASE) database is a retinal vessel reference dataset acquired from multi-ethnic school children. It contains 28 retinal images taken using a hand-held NM-200-D fundus camera having a field of view of 30 degrees.
This database
is a part of the Child Heart and Health Study in England
(CHASE), a cardiovascular health survey in 200 primary
schools in London, Birmingham, and Leicester. The part
of the investigation involving ocular imaging was carried out
in 46 schools and demonstrated associations between retinal vessel tortuosity and early risk factors for cardiovascular
disease in over 1000 British primary school children of different ethnic origin
The images in this database have a resolution of 999 X 960.
Reference: Fraz et al. (2014)
IDRiD Database
Indian Diabetic Retinopathy Image Dataset (IDRiD) database is a recently published retinal image database to evaluate the performance of algorithms developed for automatic detection and grading of DR and DME using retinal fundus images.
The database consists of 516 images (413 for training and 103 for testing), for which the manually marked OD center and fovea location are available. Also, another 81 images (54 for training and 27 for testing) are available, where the manually segmented optic disc boundary ground truth versions are available.
Images were acquired using a Kowa VX-10 alpha digital fundus camera with 50 degrees field of view, and all are centered near to the macula. The images have a resolution of 4288 X 2848 pixels and are stored in the JPEG file format.
Reference: Porwal et al. (2018)
UoA-DR Database
The University of Auckland Diabetic Retinopathy (UoA-DR) database contains 200 retinal images (56 healthy images, 114 images affected by NPDR, and 30 images affected by PDR).
These images were captured by a camera with a field of view of 45degrees and Zeiss VISUCAM 500 lenses. The images have a resolution of 2124 X 2056 saved in JPEG format.
The ONH center and boundary, macula, fovea, and the retinal vessels of all the 200 images in this database are manually segmented by a specialist ophthalmologist, who acted as the first observer and by an optometrist who was the second observer. Also, all the 200 images of this database are graded according to the severity of the DR based on the International Clinical Diabetic Retinopathy scale, by a specialist ophthalmologist.
Reference: Chalakkal et al. (2017)
EyePACS Database
EyePacs is a non-proprietary web-based application digital image system for storage and management of patient's retinal images.
EyePacs have provided the public a large dataset of retinal images from diabetic screening programs. The dataset is available from Kaggle. It consists of 35,126 images acquired with a variety of fundus cameras.
Reference: Cuadros et al. (2009)
ORIGA-light Database
The Online Retinal Fundus Image Dataset for Glaucoma Analysis and Research (ORIGA) database consists of 650 images acquired through the Singapore Malay Eye Study (SiMES).
SiMES is conducted by the Singapore Eye Research Institute (SERI).
The images were marked by experts. The dataset includes 168 glaucomatous and 482 non-glaucoma images.
Reference: Zhang et al. (2010)