I am going to extract optical cup and disc from retinal image and extracting features to classify using SVM or any other classifier. Now i want the best features to be considered in local approach.
You can use an end-to-end method using a Convolutional Neural Network. In this way, you don't need to involve yourself with feature extraction.
If you want to use something like SVM, I think it's better to extract any feature you can. Then you can apply a feature selection algorithm to select the best ones.