Recognition of Optic Disk by Utilization of Histogram of Gradients

A new method for recognition of optic disk (OD) location is presented. The proposed method is a window-based model performing on the basis of local distribution of gradients. Initially, a set of images are produced which are generated by a binary classified data including OD and non-OD. Subsequently, by partitioning of each image, weighted histogram of gradients in all eight major directions are calculated. These histograms are then placed in a sequenced order and are labeled as feature vectors. The vectors are then used as system’s input to be classified on the basis of support vector machines (SVM). From this stage on, the system through scanning the received images assigns values for classifying the findings via SVM and hence recognizes the type. In this proposed method, system does not exhibit any sensitivity toward over illumination of the images and furthermore, it’s invariant to small shifts as well as rotations. The new method s validity was measured by using 40 images having different quality, from which 37 correctly identifed the OD. This represents a value of 92.5% accuracy. The findings supersedes the exiting methods in accuracy.

Keywords:

Histogram of gradients , Support vector machine , Optic disk , Feature extraction