Fig. 5(c) and (d) show classification using the optimal projection
line. It is intuitive that by projecting object and clutter points to
an optimal line, better classification performance can be achieved.
Note that the proposed features have the capability to capitalize
on the absolute difference in the brightness between the object
class and the majority of clutter class images. For example, when
the object points lie on the top-right or bottom-left corners of the
2D SRFS, better classification can be obtained by setting optimal
line parallel to the vector 1 = 2. Traditional features, however,
can produce good classifiers only when the majority of the object
points would lie in left-top or the right-bottom of the 2D
SRFS.