A Hybrid Methods For Automatic Detection of Optic Disc in Retinal Images
Biomedical image analysis, which is becoming more important recently, is an efficient field for researchers. With the development of image processing techniques, to determine the features of the retinal images also facilitates the diagnosis of retinal diseases. There are many recent studies with the diagnosis of diseases. In this paper, we aimed to detect the optic disc(OD) accurately which is fundamental of automatic diagnose of the diabetic retinopathy disease. OD is the brightest area of the retinal image and it looks like a circle, so we have used this properties and have developed pixel-based and model-based hybrid method. For the detection of OD, we have used open access DRIVE retinal image database and for each image, we have extracted their green channels. Then, we have implemented Adaptive Histogram Equalization (AHE) to improve clarity of the images. After that, we have saved the light-color parts which is change according to the brightness level of each images by dynamically change treshold values and after the mathematical morphology processing, we have deleted the small pieces of shapes. Then we have determined the circular regions which have specific radius ranges by applying model-based Circular Hough Transform (CHT). In this way the location of the OD is determined locally. The accuracy of the study carried out on 38 images recorded as 93.37% and the average computation time of one image is measured as 1.811 seconds.
Index Terms- Image Processing, optic disc, circular hough transform, mathematical morphology, adaptive histogram equalisation