Underscan's blog/

OK, now this has been bugging me since I-don't-know-when and since here at Ipernity there are tons of people that deal with images and computing all the time I'll seek help here.

The task is really simple: filter out landscape / portrait format images from a pool of pictures.

I'd like this to happen on the command-line - if possible with out-of-the-box tools and not with special software.
If it were a cross-platform solution, that'd be great but primarily I need a Linux-based idea.

It would be nice if in the end I had a simple script that I tell a directory (possibly with subdirectories to look into recursively) with images. It "scans" these images for their height and width values, figures out if they are oriented in landscape (width > height) or portrait (height > width) or square (width = height) and prints out a list of filenames of the previously defined format.

This list could then be directed into a text-file or piped to another program etc.

Possible tools could be the "identify"-command from the great Imagemagick toolbox or maybe the "list"-option from the equally great feh image viewer.

Both are capable of figuring out the dimensions of an image.

Probably identifiy is the better choice since it allows a range of operations on the parameters of an image-readout.
It can also read out EXIF information.

[Update]
OK, maybe another possibility would be the excellent Perl library exiftool which - as its name suggests - is a Swiss-Army-knife to read/manipulate metadata.
You can specify certain tags to be read out such as ImageSize which will present you directly with the desired values.

I just learned that the identify-command has a similar option.
[Update]

[First possible solution]
for a in *; do identify -format "%f:%[fx:w/h]" $a; done

This will print out the filename (%f) followed by a colon and then the result of width divided by height (fx:w/h), e. g.
img_1654.jpg:1.5
img_1655.jpg:0.666667

It's fast and the CPU load looks reasonable even though not necessarily light.

Next steps are probably filtering out the pictures of the desired orientation (>1, <1, =1) and printing out the filenames.

I'm not sure if what I am doing here is really clever or just script-kiddie hacking.
[First possible solution]

[Quick-and-dirty solution for the moment]
Ok, now this will do for the moment but it's not really much more than a command-combination that needs to be turned into a proper script with input possibilites.

for a in *jpg; do identify -format "%[fx:w/h]%%%f" $a; done | sort | grep ^1 -v | sed 's/.*%//g'

As explained above this uses the identify-command from the ImageMagick toolbox to extract width and height of images. These values are divided and printed out in the form result%filename. This is now sorted so the portrait format pictures (< 1) are displayed first, followed by possible squares and then the landscape format ones. Doing an inverted grep of all lines with a 1 at the beginning sorts out only those in portrait layout. Using sed the numbers are eliminated and only the filenames remain.

Not really super-great but OK.

I would still be glad about input and advice. :)
[Quick-and-dirty solution for the moment]

I wonder what is the best way to go about this task, especially which is the fastest and less CPU-intense way.
Is it easier to read out the EXIF information and work with that because only the image headers need to be processed? feh - as far as I understand - loads the whole image to determine its dimensions and this takes time and calculation, esp. on large images.

I would be really thankful for ideas, hints to other people that may have challenged this idea before etc.

Thank you and - while I'm at it - merry christmas. :)