Smart Sharpening in Imagemagick

So, I’ve found that the Smart Sharpening workflow described here can give really stellar results. But I’m lazy, and was interested in a way that I could adjust images without having to open them in a graphical environment, and preferably do a bunch at once. I essentially converted the steps described by gimpguru into imagemagick and ended up with something that works. I’m still testing a few different ideas, but I find it’s working reasonably well at the moment (you can judge for yourself on the basis of examples, though the flickr images will give you a better idea). But in the spirit of sharing… you can find it here.

Timings: 10 MP image, about 35 seconds
Note this will avoid name collisions presently, so if you have a 2, 4 or 8 core cpu, and you use (ls *.jpg|xargs -n 1 -P # smartsharp.pl) you can get a 2, 4 or 8x speedup, or 20, 10 or 5 seconds per image. This is really not a lot of time, and trivial compared to the amount of time you could expect to spend in the gimp doing the same thing.

Apparently someone has already tried to implement this in the gimp. This would be a nice way to try a bunch of different parameters, unfortunately it does not have a preview (and that would be very hard to do), and parameters would not be a perfect match from that one to this one, as I use a different edge detection method. And I don’t believe I can use the gimp edge detect method without quite an investment. I’ll investigate though.

Some examples: These are crops from three larger images (full size available here), the images are the original, the final and the mask generated by the script used for sharpening,.







Advertisements

Pattern Noise Reduction

This little algorithm seems to significantly reduce pattern noise in digital images.  Instead of doing dark frame subtraction, the output value at each pixel is taken to be the average of the initial value with weight (1-dark) and the value of pixel in a gaussian blurred image with weighting of (dark).  Where dark is an element of [0,1].

I wrote this because I wanted something I could apply to webcam images both night and day that would eliminate hot pixels in night shots and would not create dark spots in day shots. Seems to work pretty well.

This is pseudo perl, and calls the imagemagick library, but you get the idea.

# pseudo-code creative commons – attribution – share alike
# author Joseph Bylund
# i.e. please copy, please give to your friends, please modify, please give me credit
use strict;
my $corrections = “pixel_corrections.png”;
my $mydark = “./dark_frames/dark_frame_avg.png”;
my $start = “webcam_pic.png”;
my $lessdark = “lessdark.png”;
my $startbydark = “startbydark.png”;
my $radius = 4;
system(”
convert -compose multiply -composite $mydark $start $startbydark
convert -evaluate Multiply 2.54 $startbydark $startbydark
convert -compose minus -composite $startbydark $start $lessdark
convert -gaussian-blur 20x$radius $lessdark $lessdark\_blur.png
convert -compose multiply -composite $mydark $lessdark\_blur.png $corrections
convert -evaluate Multiply 2.54 $corrections $corrections
convert -compose plus -composite $lessdark $corrections final.png”);

Free NEF to jpg Conversion

As in free software of course, but coincidentally it doesn’t cost you anything either. This conversion takes care of a few things,
1) raw -> jpg, the most basic aspect, we want to turn our nef files into jpg files
2) noise correction – as a function of iso, I haven’t played with the amount of noise correction yet, I just went with a function that starts at 100 at iso 100 and ends up at 1000 at iso 1600, which is the range of my camera, and according to the documentation of dcraw a normal range of noise corrections
3) it crops some pixels, my camera has some goofy edge effects, so I just cut them off
4) it preserves exif data, to me this is a pretty big one
5) it’s embarrassingly parallelized, i.e. it processes each image individually. And it will use one fewer cpu than your machine has. For instance I have an i7, with 8 cpus (after hyperthreading) so it uses 7, leaving me one to have a responsive computer during the time it’s processing images. This has the nice upside of significantly speeding up conversion, as I get approximately a 7x speedup, so processing a few hundred photos takes on the order of minutes instead of on the order of hours (1 hours 10 minutes -> 10 minutes).
6) white balance – I’m not sold on the white balance correction in dcraw, but I think it’s better than nothing. I sort of like that of the gimp, but I find that it sometimes introduces color casts. I think long term I will be moving towards camera white balance.

What it doesn’t do (yet)
1) geotagging – I think I’ve got this figured out, detailed a few posts back.
2) auto vignette correction by lens and focal length – seems like it should be pretty possible to do, I just haven’t done it yet. It would involve making a directory of averaged images at each focal length and then doing some sort of image magic compose divide. The biggest hassle would invariably be making the correction images.
3) chromatic aberration correction by lens and focal length. In a way easier, as one only needs to determine the multipliers for the different channels at different focal lengths and lenses. As opposed to having a collection of correction masks as for vignette correction. But finding the multipliers might be more difficult. I think it’s going to involve taking a picture of the sky with bars (buildings perhaps) on the edges of the frame. Essentially I’m looking for an instance that would create large chromatic aberrations. Then I can do a decompose in gimp and count pixels. Though I’m expecting the correction to be only 1-4 pixels over nearly 4000 pixels horizontal. Anyways, eventually I may try it and see what effect it has.

And finally the scripts:

#!/usr/bin/perl
#image_converter.pl
use strict;
my $numcpus = `grep -c processor /proc/cpuinfo`;
chomp($numcpus);
$numcpus -= 1; #save one processor for doing other stuffs
print "using $numcpus processors\n";
system("ls *.NEF *.nef 2>/dev/null|xargs -n 1 -P $numcpus rename 'y/A-Z/a-z/'");
system("ls *.nef|xargs -n 1 -P $numcpus chmod -x");
system("ls *.nef|xargs -n 1 -P $numcpus ~/bin/image_converter_inner.pl");
exit(0);

#!/usr/bin/perl
use strict;
my $numcpus = `grep -c processor /proc/cpuinfo`;
chomp($numcpus);
$numcpus -= 1; #save one processor for doing other stuffs
my $image = $ARGV[0];
$image =~ s/.nef//;
my $iso = `exiftool -ISO $image.nef|awk '{print \$3}'`;
chomp($iso);
my $waveletcorrect = log($iso/100)/log(2);
$waveletcorrect = 100*3.162^$waveletcorrect;
system("dcraw -t 0 -a -n $waveletcorrect $image.nef");
system("convert $image.ppm $image.jpg");
system("/bin/rm $image.ppm");
system("exiftool -q -overwrite_original -tagsFromFile $image.nef $image.jpg");
my $width = `identify -format "%[fx:w]" $image.jpg`;
my $height = `identify -format "%[fx:h]" $image.jpg`;
chomp($width);
chomp($height);
$width -= 4;
$height -= 4;
my $cmd = "convert -crop $width\\x$height+0+0 +repage $image.jpg $image.jpg.tmp; mv $image.jpg.tmp $image.jpg;";
system("$cmd");
exit(0);