object detection using neural networking

How can I input the feature vector produced by the 'extractSURFFeature' in to a neural network in order to classify / detect object in a static image. Is 'extractsurffeature' out put is really a good choice for doing so? how much accuracy can I expect on this?