Abstract:

Millions of people have problems with their vision every year. A comprehensive dilated eye exam is increasingly becoming part of a regular health check-ups. A human eye health is assessed by examining its retinal fundus image. A deviation from the normal pattern present in the image provides clues to the various ocular diseases.  The inspection was done in the past manually by an expert ophthalmologist. The need for a computer-assisted analysis of retinal image is growing with advancement of technology and increase in workload of the clinicians. The main task in a computerized retinal image analysis is the identification of the major parts of retina such as optics disc and macula; optical disk localization is the foremost important one here. Different optical disc localization algorithms have been suggested in the past based on its characteristics of being the brightest part and circular in shape. The method proposed here is based on an image contrast-stretch scheme, referred to as Speeded-Up Adaptive Contrast Enhancement (SUACE), where dynamic range for each pixel is adjusted using a single parameter to limit the amplification. It was observed that by reducing the parameter in steps from highest to lowest value, we can reach a value for parameter that makes the optic disc region (brightness part) highly discriminative as compared to the rest of the image. This helps greatly in localizing the optic disc region. The results of practical work were obtained using a publically available dataset, which achieved good accuracy with reduced computational time.