I need to train svm with N images.Feature vector to SVM is computed using SURF.Is there any way to limit the keypoints computed other than using minhessian,Octaves and octaveslayer? OR How can I compute same number of keypoints for all the input images?