834127.fig.004a
(a)
834127.fig.004b
(b)
834127.fig.004c
(c)
Figure 4: Abs. training set size versus test set accuracy for pixel-based features on MNIST, USPS, and DIGITS (SVM variants).