Convolution is a very common operation during data analysis and is thoroughly described as part of Gnuastro’s Convolve program which is fully devoted to this job. Because of the complete introduction that was presented there, we will directly skip onto the currently available convolution functions in Gnuastro’s library.
As of this version, only spatial domain convolution is available in Gnuastro’s libraries. We have not had the time to liberate the frequency domain function convolution and de-convolution functions that are available in the Convolve program267.
Convolve the given
tiles dataset (possibly a list of tiles, see
gal_data_t and Tessellation library (tile.h)) with
numthreads threads. When
edgecorrection is non-zero, it
will correct for the edge dimming effects as discussed in Edges in the spatial domain.
tiles can be a single/complete dataset, but in that case the speed
will be very slow. Therefore, for larger images, it is recommended to give
a list of tiles covering a dataset. To create a tessellation that fully
covers an input image, you may use
gal_tile_full_two_layers to also define channels over your input
dataset. These functions are discussed in Tile grid. You may then
pass the list of tiles to this function. This is the recommended way to
call this function because spatial domain convolution is slow and breaking
the job into many small tiles and working on simultaneously on several
threads can greatly speed up the processing.
If the tiles are defined within a channel (a larger tile), by default
convolution will be done within the channel, so pixels on the edge of a
channel will not be affected by their neighbors that are in another
channel. See Tessellation for the necessity of channels in
astronomical data analysis. This behavior may be disabled when
convoverch is non-zero. In this case, it will ignore channel borders
(if they exist) and mix all pixels that cover the kernel within the
Correct the edges of channels in an already convolved image when it was
initially convolved with
convoverch==0. In that case, strong boundaries might exist on the
channel edges. So if you later need to remove those boundaries at later
steps of your processing, you can call this function. It will only do
convolution on the tiles that are near the edge and were effected by the
channel borders. Other pixels in the image will not be touched. Hence, it
is much faster.
Hence any help would be greatly appreciated.