Article Preview
TopNumerous methods have been devised to address the segmentation of blood vessels in retinal images. (Hoover, Kouznetsova, & Goldbaum, 2000) proposed a method to segment the blood vessel by using local and region based properties at each pixel to detect the vascular tree. Pixels are classified as vessel or non-vessel by thresholding the image generated by a matched filter using a probing technique. Probing allows a pixel to be tested in multiple region configurations, before the final classification.
(Staal, Abràmoff, Niemeijer, Viergever, & Van Ginneken, 2004) used the notion of ridges for extracting the blood vessel. The properties of the ridges and the pixel considered forms the feature vector. Then K-nearest neighbor (KNN) classification was adopted for classifying the image ridges.
The method (Soares, Leandro, Cesar, Jelinek, & Cree, 2006) presented uses supervised classification. Each image pixel is classified as vessel or non-vessel using the pixel feature vector, which is composed of the pixel intensity and 2-D Gabor wavelet transform responses taken at multiple scales. Then a Bayesian classifier is applied to obtain the final segmentation.
(Singh, Kumar, & Srivastava, 2015) proposed an automatic local entropy thresholding retinal blood vessels segmentation method by modifying the standard Gaussian shaped matched filter. This method uses adaptive local thresholding to produce a binary image, then extract large connected components as large vessels (L. Xu & Luo, 2010).