Feature extraction and classification of retinal images for automated detection of Diabetic Retinopathy

Detection of Diabetic Retinopathy: A Survey

“…Match filter and first-order derivation of Gaussian matched filter technique are applied in [8]. Fuzzy C-Means clustering and morphological operations are used to detect blood vessels [10]. The Retinal fundus image is converted into a CMY color image after that isolate Magenta channel [11].…”

“…A. Pre-processing Pre-processing is an important step in process of detecting DR. Most of the researchers convert retinal fundus image into green channel [1][2][3], [5][6][7][8], [10][11][12][13][14] as a contrast of MAs and HMs are high in this channel. Contrast limited adaptive histogram equalization (CLAHE) is applied to enhance the image and then median filter to remove noise [4, 6, 7 and 10].…”

“…To detect EXs, we need to detect a portion of the optic disc in retinal fundus image because Optic disc and EXs both have same intensity [7][8]. Optic disc mark was created using brightest pixels in an image and its coordinates are a center of the marks [2,10]. Three facts are used to detect optic disc [6].…”

“…Pre-processing techniques followed by extraction of features like blood vessels, MAs, optic disc, EXs and then applied SVM classification [10]. Enrique et al [11] applied same flow chart as [10] but used different methods for extraction and used two classifiers SVM and decision tree. After applying pre-processing technique feature extraction, feature selection using Haar wavelets and principal component analysis is done and then applied classification techniques like OneR and BPNN [12].…”

“…MAs and EXs both are extracted using Selective Binary and Gaussian Filtering regularized Level Set (SBGFRLS) on pre-processed fundus images and matched filter with the first-order derivative of Gaussian (MF-FDOG) is used to detect blood vessels [8]. Pre-processing techniques followed by extraction of features like blood vessels, MAs, optic disc, EXs and then applied SVM classification [10]. Enrique et al [11] applied same flow chart as [10] but used different methods for extraction and used two classifiers SVM and decision tree.…”