
details (high-frequency components) are computed by convolving c
jþ1
with h
0
(n)
multiple times and then once with h
1
(n). Since large structures are composed of
low-frequency components, dividing the spectrum into subbands actually sep-
arates different scaling components in the signal. The concept of spectrum
division greatly facilitates understanding of filter banks.
In practice, signals and images are recorded in discrete and finite form.
At each decomposition, the number of coefficients in c
j
(n) becomes half the
number of samples in c
jþ1
(n). Thus, the decomposition can repeat until only one
approximation coefficient is left. The maximum number of decomposition ...