assume that I have a set of image and segmented image of a object in image (training set). Now, I want to measure features of object in image based on the training data. But I don't know how many features in image. And which is best feature for image segmentation? Gradient feature or entropy feature or ....Could you explain help me ?