Find images in a Linux directory based on their resolution

Question

I would like to scan all images in a directory (recursively within sub-folders), and find those with resolution higher than a specific threshold (e.g. say those with resolutions at least 800x600 or if easier, say with width higher than 1000 pixels). Then I would like to log their address in a text file, accompanying their resolution (or [width], [height] for a better formatting).

So log.txt would look like this:

/home/users/myuser/test/image1.jpg, 1800, 1600
/home/users/myuser/test/image20.jpg, 2800, 2600
/home/users/myuser/test/image30.jpg, 1500, 1200

How can I do that that using a bash script? I have to scan millions of images.

Solution

Via bash's recursive glob and ImageMagick's identify command:

shopt -s globstar
identify -format "%f, %w, %h\n" **/*.{png,jpg,jpeg}

Saving such output to file , is just a matter of adding > mylog.txt to previous command, that is

identify -format "%f, %w, %h\n" **/*.{png,jpg,jpeg} > mylog.txt

From there, you could use awk or perl to compare mylog.txt columns

awk -F ',' '$2 > 800 && $3 > 600' mylog.txt

awk here uses , as separator for columns, and the usual structure for awk is /PATTERN/{COMMANDS}, which defaults to just printing if {COMMANDS} omitted ; in the particular example above, if the pattern $2 > 800 && $3 > 600 is true, that is it's the image above desired resolution, you'll get it printed to the screen.

And probably skipping the log step in between, it would be a little better to just pipe everything:

shopt -s globstar
identify -format "%f, %w, %h\n" **/*.{png,jpg,jpeg} | awk -F ',' '$2 > 800 && $3 > 600' > filtered_images.txt

---

In case you encounter arguments list too long error, typically find command is better approach for recursively walking the directory tree. The identify can be called through find's -exec flag, and filtering still can be handled by awk:

$ find -type f -regex "^.*\.\(png\|jpg\|jpeg\)$" -exec identify -format "%f, %w, %h\n" {} \; | awk -F ',' '$2 > 800 && $3 > 600' 
fanart.jpg, 1920, 1080
fanart.jpg, 1920, 1080
globalsearch-background.jpg, 1920, 1080
fanart.jpg, 1280, 720

As usual, don't forget to add > log2.txt to save everything to file.

Full path of to the file could be handled in either one of two ways. One, by specifying %d/%f in identify command's format string, or use find's -printf option. That is either

find -type f -regex "^.*\.\(png\|jpg\|jpeg\)$" -exec identify -format "%d/%f, %w, %h\n" {} \; | awk -F ',' '$2 > 800 && $3 > 600'

Or

find -type f -regex "^.*\.\(png\|jpg\|jpeg\)$" -printf "%p, " -exec identify -format "%w, %h\n" {} \; | awk -F ',' '$2 > 800 && $3 > 600'

Comments

[fiatux]

also asked on superuser: superuser.com/q/1333517/4714

[Tina J]

So basically you propose to first log all files (one log.txt here), then apply the comparison filter and have a second log2.txt. Right?

[Sergiy Kolodyazhnyy]

@TinaJ Yep, or as I just added to my answer, you can just pipe everything from identify to awk, and save on creating the log1.txt in between.

[Tina J]

Yeah that would be better as it will take a long time. I guess you are missing > log.txt at the very end of your pipe command?!

[Sergiy Kolodyazhnyy]

@TinaJ Not anymore :) Edited to include that as well

[Tina J]

I get -bash: /usr/bin/identify: Argument list too long

[Tina J]

I guess there are too many results for identify to scan!

[Sergiy Kolodyazhnyy]

@TinaJ How so ? Are you getting argument list too long error ?

[Tina J]

yeah I wonder. I run it on a folder with subfolders, containing around 75K images.

[Sergiy Kolodyazhnyy]

@TinaJ Alright. I'll add a slightly different approach then in a minute, alright ?

[Sergiy Kolodyazhnyy]

@TinaJ See the edit :)

[Tina J]

ok I trust it would work! Will test it later. Tnx!

[Sergiy Kolodyazhnyy]

@TinaJ Very welcome :)

[Tina J]

Sergiy, one small tweak is that right now it only shows the file name, not with the file's path. Can you make this small change? Since in my original question, address is being asked!

[Sergiy Kolodyazhnyy]

@TinaJ Edited the answer as requested. It can be done via two methods.

[Tina J]

Cool! Looks great.