|
In most dual-band OCT systems, there is a spectral gap between both bands. This might be as large as one third of the
total spectral region. Therefore, a simple Fourier transformation of the data does not give the resolution that could be
possible considering the overall spectral width. Instead, the full width of the peak is comparable to the width resulting
from a single band and is additionally modulated. We compare several methods to achieve a high resolution in spite of
the missing data. Because in a dual-band system the image quality resulting from the full information is not known we
test and optimize different algorithms by using the data from a single band system and excluding an equal part of the
spectrum. While methods using non-equidistant sample points like Vandermonde and Lomb transformation work well
with small spectral gaps, they result in large image artifacts for broader gaps, which are typical for dual-band OCT
systems. Simulations show that fitting the available data with a limited set of sine and cosines functions might give good
results but for larger gaps and appropriate amount of basis-functions this method fails, too. Dividing both bands into
overlapping smaller bands and looking at the phase of short-time Fourier transformations (STFT) resulting from a single
scatterer, it becomes clear that the amplitude of all Fourier coefficients for the total band can be estimated by the change
of the phase of the STFTs in and between both bands. Therefore, we developed an algorithm of weighting the data based
on the phase distribution of the STFT data. As a single value specifying the phase distribution we choose the absolute
sum of the STFTs divided by the sum of the amplitudes of the STFTs. Because typical OCT data are not caused by
single scatterers, we adapted this algorithm with a cluster analysis to predict the appropriate amplitude expected for a full
spectrum from the phase distribution of the STFTs inside both bands and between both bands. Although the image is
noisier and fainter compared to an image from the full spectrum, the resulting image has the best resolution from all
methods investigated.
|
