Abstract:
|
In this work, a comparative study between Elliptic
Fourier and B-spline descriptors is carried out for comparing
their efficiency in characterizing the contour shape of image
objects. In both cases, the goal is to obtain the least
representation error using the fewest possible number of
coefficients. With Fourier descriptors, different number of
harmonics are used while the remaining ones are set to zero. In
the B-spline case, coefficients are obtained iteratively using a
least-square filter, followed by a decimation procedure. Linear
and cubic B-splines are considered. In general, data will be more
compressed when the lower number of coefficients is used, but
the representation error also increases considerably. We use a
signal/error ratio, expressed in dBs, to measure the similarity of
each approximation. The signal value is obtained from the
‘modulo’ addition of all coordinate points, whereas the error
value is computed accumulating the ‘modulo’ distance between
original and reconstructed shape. It can be shown that for a
lower compression rate, the results do not vary significantly in all
three methods. For higher compression rates, Elliptic Fourier
Descriptors are more efficient than linear and cubic B-splines,
especially in soft contours, but B-splines have lower
computational cost. |