Affine-invariant modeling of shape-appearance images applied on sign language handshape classification