multi class support vector machine for hand gesture recognition

I need to clarify a question related to sift algorithm and support vector machine. Imagine I have images of hand signs. From sift http://www.cs.ubc.ca/~lowe/keypoints/ i'll get the descriptors of every image. Can this descriptor be used as feature vector to train the multiclass support vector machine in matlab. Or is there any way to extract most informative feature vector from descriptor for image comparison since descriptor is very huge. Thanks.