1249 Fast GLM and moment-preserving thresholding for retinal vascular segmentation and denoising KHAN BAHADAR KHAN 1 *, AMIR A.KHALIQ 1 , MUHAMMAD SHAHID 2 1 Department of Electronic Engineering, International Islamic University 2 Department of Electrical Engineering, Capital University of Science and Technology Islamabad 44000, Pakistan. E-mail: {Bahadar.phdee46; m.amir}@iiu.edu.pk, shahid.eyecom@gmail.com Precise retinal vessels localization is an important and challenging task. Segmentation of retinal blood vessels become more difficult in abnormal images with the presence of diseases like hypertension, diabetes, stroke and other vascular disorders. In this work, a new fast framework for automatic retinal blood vessels extraction and denoising has been proposed. Green channel due to its prominent vessel structure is used as an input to mor- phological filters to eradicate low frequency noise or geometrical entities, e.g., macula, optic disk and other abnormalities. The Generalized Linear Model (GLM) regression is used for non-uniform contrast enhancement followed by Frangi filter for vasculature based enhancement. Masking operation has performed to extract Region of Interest (ROI) for application of moment-preserving thresholding to separate vessel and background pixels. Finally, postprocessing steps are applied to eliminate unconnected pixels and to obtain final binary image. This technique is validated on the DRIVE and the STARE databases and contested with other competing techniques. Experimental results indicate that the pro- posed vessel extraction framework outperforms many recent existing methods published in literature. Keywords: DRIVE, denoising, moment-preserving thresholding, retinal Images, STARE, vessel segmentation. 1. INTRODUCTION The retinal blood vessel dissection is deliberated as a prerequisite in many medical applications related to eye disorders diagnoses and surgery planning. It can be applied for analysis of numerous eye abnormalities such as glaucoma [1], Diabetic Retinopathy (DR) [2], [3] and hypertension [4], etc. Moreover, it can be also utilized for examination of ves- sel features such as vessel width and tortuosity. Retinal abnormalities that change the vas- cular tree caused visual impairment. Detection of such abnormalities at the primary stage can restrain the blindness to a great extent. Hence, retinal blood vessel extraction and fur- ther investigation of its features help in diagnosis of retinal diseases [5]. Retinal vascular manual extraction is very hard and time taken task that needs expertise. Therefore, the evolution of an automatic techniques for vessel extraction and vessel width computation is required [6]. Nowadays, digital ophthalmoscopes make it easier to acquire retina digital images. This helps the medical doctors to perform retina vasculature analysis of their patients and make it possible to develop automated image investigation techniques. Various methodol- ogies have already been published for retinal vasculature extraction. However, automatic extraction of retinal vessel map is also challenging because of its different structure com- plications and impact from other sources. The complexities originate due to vessel contrast