Article Preview
Top1. Introduction
Segmentation of retinal blood vessels to extract retinal information is used to predict diseases (Fong et al., 2004) such as ophthalmic disorder (Fraz et al., 2012), obesity (Wang et al., 2006), hypertension (Foracchia et al., 2001), glaucoma (Mitchell et al., 2005), and diabetic retinopathy (Goatman et al., 2011). The Retinal images are obtained from a fundus camera for diagnoses of retinal diseases. The captured retinal image is low contrast due to the low quality of the fundus camera. However, it is very close to the background and vessels of retinal images; therefore, it is very difficult to predict retinal diseases; therefore, to solve this issue of accurate extraction vessel structure, we proposed the segmentation process method to provide the best result of matched filtered approach compared with other matched filtered approaches.
The matched filter (MF) kernel-based approach is used to extract the intensity profile of vessel structure. The kernel is designed based on the vessel's piecewise line segment's limited curvature and the vessel's width gradually diminishing toward the optic disk.
In literature, the first matched filter-based kernel design by author Chaudhari et al. (Chaudhuri et al., 1989). They state that the shape of the curve of the cross-sectional intensity profile is Gaussian. Most researchers in literature surveys improve the matched filter approaches by using the thresholding technique. The blood vessels' shape, location, and scale feature the matching filter kernel based on parameters.
However, Author Zolfagharnasa (Zolfagharnasab & Naghsh-Nilchi, 2014) improved the gaussian pdf by using the Cauchy distribution-based matched filter as a proposed approach. They state that the retinal image of the intensity profile of the Cauchy curve. But they are not accurately extracted blood vessels. As a result, Singh et al. (Singh & Singh, 2020; Singh & Srivastava, 2016a) proposed a Gumbel pdf-based matched filter to improve the performance of extraction of retinal blood vessels. They claim that the best parameter value is used to produce an MFR image to extract blood vessels based on optimal thresholding-based entropy. However, to improve the accuracy of accurate extraction of blood vessels. Therefore, we proposed a novel approach of Generalized Gamma pdf, which outperforms existing approaches in terms of accuracy.
In this paper, the proposed matched filtered is designed the Generalized Gamma pdf for analysis of the intensity profile of blood vessel and examine the Generalized Gamma pdf with another pdf curve; we analyze that the proposed method of Generalized Gamma pdf curve is a much better result comparing with various probability distribution function curves are shown in figure 1. The proposed model is presented in Section 2, the performance evaluation in section 3, the result analysis in Section 4, and the conclusion in Section 5. The following section discusses the rest of the paper.
Figure 1. Analysis of the Gaussian, Cauchy, Freshet, and Generalized Gamma curves
Top2. Proposed Method And Model
The Generalized Gamma pdf kernel's proposed model extracts accurate blood vessels. The model is a design based on a combination of three steps. They are preprocessing steps, Generalized Gamma pdf kernel design, and postprocessing containing optimal thresholding, length filtering, and masking.