Research On Optic Disc Detiection In Colour Retinal Images

Digital fundus image have played very important role both in medical research orclinical diagnosis and treatment. It can not only be used for the diagnosis of fundusdisease, but also provide the reference basis for the diagnosis and treatment ofhypertension, diabetes, stroke and other systemic diseases. It is objective, accurate andless time-consuming to use digital image processing technology for enhancingfeatures and obtaining morphological parameters in retinal image. Digital imageprocessing technology have provided a more reliable basis for clinical diagnosis. Inthe computer-assisted diagnostics system, optic disc localization is the prerequisiteand key step of clinical diagnosis and fundus structure analysis.In a healthy retinal image, the features of optic disc are obvious. For instance, thebrightness of optic disc is higher than other structures. Retinal vessels gather at thedisc center. And the direction of the vessels stretching out from disc are very similar.Although the optic disc has well defined features, automatic optic disc localization isnot easy. This is due to many distracters, including nonuniform illumination,pathologies (exudative lesions, peripapillary atrophy), vessel edges, individualdifference, etc. Thus, recognition of optic disc is the object in our research. It includeslocalization and segmentation of optic disc.According to the features of optic disc, a locating algorithm based on templatematching was implemented. And then, a new algorithm based on mapping has beenpresented. The new method is easy to implement and accuracy. Four public fundusimage datasets were used to evaluate the performance of the proposed method. Andthe total success rate is96.6%with an average computing time of1.3s for each image.The algorithm is less time-consuming and high accuracy compared to theart-of-the-state location method.In the phase of segmentation, the algorithm based on hough-circle transform wasimproved. Because of extracting the region of interest and removing vessels,hough-circle transform was refined. Due to the edge of the optic disc gray gradientlarger and approximate ellipse, a method using ellipse fitting has been developed. Themethod is precise and fast. the total success rate is84.5%with an average computingtime of1s for each image.