GA based Automatic Optic Disc Detection from Fundus Image using Blue Channel and Green Channel Information

Abstract

Proper detection of Optic disc (OD) in a fundus image is an important stage in an automated method to screen diabetic retinopathy. To identify the OD the intensity of the pixel is used. The pixel of the OD is different from the rest of the fundus image and intensity based techniques such as thresholding may be applied for the detection. Due to more brighter region false resemblens of OD is generated and the intensity based technique fails to detect the OD properly. Using techniques like template matching the OD is detected [1] and [2]. The sliding window tehnique applied in template matching is time consuming. The method proposed is using Genetic Algorithm which will search the entire fundus image in a short time. Almost in every intensity based method, the red channel and green channel informations are used. The Blue channel information is neglated and the reason for avoiding is no potential information in the blue channel, which is not true. Since we are detecting the OD and not the Exudates or other symptoms the blue channel will be suitable for a dull image.In this work, we tried to use the blue layer information as the major clue for the successful detection of optic disc location and size in a typical fundus image. We designed a genetic algorithm based algorithm as well as a direct search based algorithm to locate the exact position and size of the OD using blue layer and green layer intensity in a suitable fitness function.