A ROTATION INVARIANT RETINAL VESSEL EXTRACTION APPROACH USING A NOVEL IMAGE ENHANCEMENT TECHNIQUE COUPLED WITH HIGH BOOST FILTERING

Abstract

Retinal Disease analysis using fundus photography needs the expertise from engineering domain. Using fundus photographs provides a non-invasive way to inspect the retina of human eye and in turn the window for human vasculature. Retina of human eye gets affected with abnormalities as a result of diabetic and hypertensive retinopathy. In order to recognize the abnormalities and grade the severity of disease, it is indeed required to locate the retinal landmarks accurately. Exact identification of Retinal vasculature in terms of position and diameter is imperative. A computationally simple, robust, effectively extracting tiny vessels and rotation invariant vessel identifier algorithm is proposed. The process is carried out in four steps — (i) A novel morphological contrast enhancement — high boost filtering (MCE-HBF) method is used for gray scale version of input fundus image. (ii) Extraction and the dark objects enhancement are carried out using high boost filtering. (iii) A single optimal threshold is used for segmentation of the dark objects. (iv) Non-vascular objects and noise are removed and a clear vessel tree is extracted at the end. The proposed algorithm has reported a sensitivity of 0.82, 0.82, 0.81; specificity of 0.99, 0.97, 0.98; accuracy of 0.97, 0.96, 0.97; precision of 0.87, 0.64, 0.81 and false positive rate of 0.0081, 0.0218, 0.010 on HRF healthy, HRF diabetic and DRIVE dataset images, respectively. With the lowest possible false positive rate, the proposed approach offers a scale and rotation invariant method for extracting minute blood vessels. The algorithm has shown better results than the state-of-the-art outcomes when tested on the HRF and DRIVE Datasets. The photos from the DiaretDB1 dataset show that the results are also useful.