Gamma ISP: Reference Manual

adf

NAME
adf - Adaptive filtering with filtering function based on local fringe spectrum.

SYNOPSIS

adf <int> <sm> <cc> <width> <alpha> [nfft] [cc_win] [step] [ymin] [nlines] [wfrac]

EXAMPLE

adf 1352_1610.flt 1352_1610.sm 1352_1610.sm_cc 2500 0.4 32 5 4 0 0 0.8

DESCRIPTION
adf reads the complex valued interferogram, computes locally the interferogram power spectrum, designs a filter based on the power spectrum, filters the interferogram , estimates the phase noise coherence value for the filtered interferogram and writes out the filtered interferogram and coherence map.  The algorithm is based upon the paper by Goldstein and Werner (Geophysical Research Letters, vol. 25, no. 21. pp. 4035-4038, 1998).

The goal of the adaptive filtering step is to reduce phase noise thereby reducing the number of residues.A residue is a point in the  interferogram where the sum of the phase differences between pixels around a closed path is not 0.0. The phase differences in the wrapped interferogram lie between -PI and +PI . Phase unwrapping determines the number of multiples of 2PI to add to every point in the interferogram based on some minimization criterium. Residues exist as pairs. Generally, thermal noise causes pairs of residues that are close together to be generated. Filtering of the interferogram reduces the noise and hence the total number of residues.

On the other hand the user must be aware that too strong filtering may introduce phase unwrapping errors by eliminating fringes. The filter significantly reduces the phase noise for areas with sufficiently reliable phase information. while high phase noise areas are minimally affected. The exponent for non-linear filtering determines the filter strength with larger values leading to greater filtering. Filtering is implemented using the FFT, which is nominally restricted to a power of 2, though use of the FFTW subroutine library, loosens this restriction, such that the FFT size for this application should be even.  For the filtering of rapidly changing phase aassociated with the  topographic phase of rugged terrain a small FFT size (16,32) is better.  Larger FFT window sizes are preferred in regions of low fringe visibility but this only when the phase slope can be assumed constant over the estimation window. In the case rather flat terrain and for terrain phase removed differential interferograms larger window sizes as 32 -->246 lead to good results. Multiple application of the filter is possible, and can lead to better results when using large values of the filter exponent parameter.

The argument of the complex valued output filtered interferogram corresponds to the filtered phase, the magnitude to a "phase noise coherence estimate" of the output image. This "phase noise coherence estimate" is a measure for the phase noise of the filtered interferogram and can be used in the phase unwrapping to mask low coherence areas. To facilitate this it is possible to write the "phase noise coherence estimate" out to a seperate file.

The processing step parameter influences the processing time. Small values lead to a slightly smoother filtered output image and longer processing times. The dimensions of the smoothed interferogram file are identical to that of the interferogram file. Therefore, the programs to generate SUN raster or BMP format images and screen display may use the same parameters.

An important parameter is the fraction of points that are non-zero within the filter window, wfrac_min. This threshold prevents filtering of regions with few valid data values such at image edges. Setting the value to .25 gives filtered output at every input pixel with non-zero input value

SEE ALSO
adapt_filt, typedef_ISP.h.