Hi Aurelien,
as far as I analyzed the source code of astrometry.net, the only program
that is really needed is indeed imcopy.
To help you out here, I attach the manpage for imcopy (I had to write
manpages for astrometry.net anyway...).
Best regards
Ole
.TH imcopy "1" "April 2014" "CFITSIO version 3.370" "User Commands"
.SH NAME
imcopy \- copies FITS images.
.SH SYNOPSIS
.B imcopy
\fI\,inputImage outputImage\/\fR[\fI\,compress\/\fR]
.SH DESCRIPTION
Copy an input image to an output image, optionally compressing
or uncompressing the image in the process. If the [compress]
qualifier is appended to the output file name then the input image
will be compressed using the tile\-compressed format. In this format,
the image is divided into rectangular tiles and each tile of pixels
is compressed and stored in a variable\-length row of a binary table.
If the [compress] qualifier is omitted, and the input image is
in tile\-compressed format, then the output image will be uncompressed.
.PP
If an extension name or number is appended to the input file name,
enclosed in square brackets, then only that single extension will be
copied to the output file. Otherwise, every extension in the input file
will be processed in turn and copied to the output file.
.PP
The full syntax of the compression qualifier is:
.IP
[compress ALGORITHM TDIM1,TDIM2,...; NOISE_BITS]
.PP
where the allowed ALGORITHM values are Rice, GZIP, PLIO,
and TDIMn is the size of the compression tile in each dimension,
and NOISE_BITS = 1, 2, 3, or 4 and controls the amount of noise
suppression when compressing floating point images.
.PP
Note that it may be necessary to enclose the file names
in single quote characters on the Unix command line.
.SH EXAMPLES
1) imcopy image.fit 'cimage.fit[compress]'
.IP
This compresses the input image using the default parameters, i.e.,
using the Rice compression algorithm and using row by row tiles.
.PP
2) imcopy cimage.fit image2.fit
.IP
This uncompress the image created in the first example.
image2.fit should be identical to image.fit if the image
has an integer datatype. There will be small differences
in the pixel values if it is a floating point image.
.PP
3) imcopy image.fit 'cimage.fit[compress GZIP 100,100;4]'
.IP
This compresses the input image using the following parameters:
.IP
GZIP compression algorithm;
100 X 100 pixel compression tiles;
noise_bits = 4 (only used with floating point images)
.SH SEE ALSO
\fBfitscopy\fP(1).
