Image destriping

Image destriping with Fast Fourier Transform (FFT)

In some image maps, artifacts (stripes) are visible. These can be reduced with FFT filtering (i.fft).

First we take a look at the original tile (screenshot upper right map):

# set computational region to GDEM map:
g.region rast=imagemap -p
d.mon x0
d.rast imagemap

Then we perform Fast Fourier Transform (FFT; screenshot upper left map):

i.fft in=imagemap real=imagemap_real imag=imagemap_imag
d.mon x1
d.rast imagemap_imag
d.rast imagemap_real

Create a MASK in order to remove noise (screenshot upper left map):

# digitize filter mask to reduce noise
r.digit filter #not working in Grass v7
d.rast filter
r.mask -i filter
d.rast imagemap_real

Inverse FFT to get filtered GDEM back (screenshot lower left map):

i.ifft real=imagemap_real imag=imagemap_imag output=imagemap_filtered
r.mask -r
d.mon x2
d.rast imagemap_filtered
r.colors imagemap_filtered color=elevation 
d.rast imagemap_filtered

Calculate the differences (screenshot lower right diagram):

# validate with differences map:
r.mapcalc "diff = imagemap - imagemap_filtered"
r.colors diff color=differences
d.mon x3
d.rast.leg diff

# done

Workflow of image destriping with FFT

Note: If the MASK is made larger (more to the center), more aggressive destriping occurs.

See also

• https://www.mapbox.com/blog/debanding-the-world/ (about Landsat7 SLC off debanding)
• Image Filtering in the Frequency Domain
• Landsat 7 destriping