Algorithm & Application Based Retinal Image Processing

Objective: The retina is one of the most important structure in eyes, and it can convert the external light into the information that brain can recognize, then, upload the information to the brain by the optic nerve, so that we can get information from the outside. Retinal image is color image of the retina through the special camera, and the important basis for reflecting the retina health. Retinal images are widely used in clinical diagnosis of diseases of the retina, and, are of great importance to the retinal image processing and analysis. Optic disk and retinal blood vessels are the main structure of the retinal image, the automatic detection and segmentation as the basic work in retinal automated processing, aiso play a certain role to the diagnosis of retinal disease.Methods: Now, the existing optic disc detection methods with high algorithm efficiency, but the low success rate at the same time; methods with high success rate, but the algorithm efficiency is low. So, the article puts forward the optic disc detection method based on Ada Boost algorithm. The method eliminate unnecessary data characteristics by changing the data structure, and, complete the detection of optic disc only rely on brightness characteristics of optic disc itself. Current unsupervised retinal blood vessels segmentation methods have certain limitation to pathological retinal image, and the accuracy of supervision retinal blood vessels segmentation method is low. So, the article proposed retinal blood vessels segmentation method based on ELM algorithm. The method extract feature according to the structural characteristics of the blood vessels on the pixel level, and then select the ELM algorithm obtaining classified forecast function, complete retinal blood vessels segmentation.Results: The average success rate, pixel distance average and the average computation time of optic disc detection in testing set is respectively 97.8%, 29.5 pixels and 1.04 seconds. Average of SEN, SPE and ACC of retinal blood vessels segmentation method in Drive database is respectively 0.9213, 0.9153 and 0.9357. In the Stare database, average is respectively 0.8267, 0.8998, and 0.9147.Conclusion: Compared with the existing optic disc detection methods, the proposed method has obvious advantages in success rate and computing time.The SEN value of the proposed retinal blood vessels segmentation method is higher than result of artificial markers, but the ACC value needs to be improved.