hpr-knowledge-base/hpr_transcripts/hpr3564.txt

Episode: 3564
Title: HPR3564: Removing EXIF data from an image
Source: https://hub.hackerpublicradio.org/ccdn.php?filename=/eps/hpr3564/hpr3564.mp3
Transcribed: 2025-10-25 01:27:24

---

This is Hacker Public Radio Episode 3564 for Thursday the 31st of March 2022.
Today's show is entitled, Removing the XIF Data from an Image.
It is hosted by Dave Morris and is about 12 minutes long.
It carries an explicit flag.
The summary is, an image might reveal data you want to keep private.
Hello everybody, this is Dave Morris.
Welcome to Hacker Public Radio.
Now today I'm just recounting some experiences I had in the past couple of weeks,
where I'm in the process of writing a script.
It's a bash script and I want to be able to deal with pictures sent in with HBR shows.
And one of the things that I'm doing in this script is to strip the XIF and other metadata from images.
And that's because this metadata can contain details that can identify the creative image.
So it usually shows the camera type and all sorts of settings on the camera.
But if you've got GPS set, it's certainly there are physical cameras that will do this.
Got GPS and they will put GPS details into images.
Not sure about phones whether that's an option.
But anyway, it seems like it's an courtesy to people sending in pictures to anonymize them as much as possible.
So as I was doing this, I discovered that just ripping out the XIF data is not the best way to go.
I haven't been doing that from some of my own pictures in the past, but there's one setting in the XIF standard.
And I've got a reference in my notes to all of the details of XIF if you want to dig deeper.
There's one setting called orientation.
If you remove that, then you can end up with a picture that's wrongly,
wrongly positioned, wrongly oriented.
So what happens is sometimes images are created with a particular orientation by the camera.
Sometimes because the camera, if you rotate it to do a portrait style image,
or you change it to a landscape setting, it takes the picture with a basic setting
like it's always in portrait or it's always in landscape.
And it saves data that says this is the camera doing it.
It saves data that says, when you display this, rotate it in some way.
Okay, so if you remove that, then your image comes out wrong.
Doesn't happen very often, but it definitely does happen.
So I'm just doing a quick, whole quick anyway, it's difficult to do quick episodes.
I can't seem to do it with Ram long too much.
But I wanted to do a quick episode about what I found, what tools I used, and how I solved this.
Just because I thought it might be interesting, and also in need of shows.
So the actual problem I was trying to solve was in early March, 2022,
somebody sent in three images with orientation values in the XIF metadata.
They'd been taken in one orientation, but were being rotated for viewing, like I was just explaining.
And I later discovered that the orientation setting can be easily viewed
with the XIF tool command, which I happen to have installed.
Not sure whether it's a default or many systems, but I had been using it on and off to strip stuff, actually.
I've given an example of XIF tool, hyphen orientation, and then the name of the file.
And you get back this particular case orientation, and then it says rotate 90 CW.
I'm not sure what CW means, but
it's to rotate the image when it's displayed.
I think that most image viewing stuff will just do the rotation that's been asked for.
And definitely if you display it on the web with that metadata in, it will rotate.
So what's happened here is the image is rotated anti-clockwise.
So the top of the image is on the left of the display picture
when you display it with that orientation.
And this needs to be reversed.
So the orientation setting says rotate 90 degrees to...
Oh, CW be clockwise, sorry, bit dense.
Anyway, rotate 90 degrees clockwise, and it's displayed with the top at the top.
It's actually stored as a number, and this is number 6.
And in the XIF tool documentation is a list of all the different orientations 1 to 8,
which do various things.
It's actually a bit more than orientation is also flipping by doing a mirror image horizontally or vertically or whatever.
So you can see that removing all the XIF data just causes an image to revert to its raw state as I have already mentioned.
And then there's a demonstration coming up in a minute.
So it told me a little while to understand this problem,
because I couldn't find anywhere anything that said this is what images contain,
and this is what the XIF data tells you.
I guess it makes sense when you think about it, maybe I wasn't thinking about it enough.
But along the way, if I try to find it more, I found a repository on GitHub,
which I mentioned in the notes, where you could just feed it a picture,
and it would generate all of the 8 possible XIF orientations from it.
And I took an old picture I did in a show way back, making coffee using a lock-apop.
And I used this tool to generate all the different picture types.
And in the notes, there's the original picture, which is actually a thumbnail,
because it's quite a big picture, but whatever.
It shows it with the orientation of 6, that is the raw tape 90 degrees clockwise.
And then I stripped the XIF data from it.
And you can see that that same picture has got the top to the left of the image.
So how would you fix this? How would you do this in a reliable way? It was my question.
I found various things in a number of searches, but you know what Google's like?
It takes you to a bazillion things, unless you ask precisely the right question.
I came up with three tools that I actually installed, called JPEG XIF Orient,
which reels or writes the exit orientation tag, XIF Auto-Tran, which transforms XIF files
so that your orientation becomes one, that is horizontal.
And JPEG-Tran, which does lossless transformation of JPEG files.
But these are only for JPEG files.
Yeah, okay, the ones I was dealing with were JPEGs.
But what if somebody sends in a PNG or the format?
So the XIF Auto-Tran tool is actually a shell script, the other two are binaries.
And it uses the JPEG-XIF Orient to find the orientation
and JPEG-Tran to undo whatever rotation or other transformation has been defined.
So it actually twiddles with the pixels of the picture.
So I had a good look at the script, the shell script, XIF Auto-Tran.
And it got a better idea of what was going on here.
So in the script I'd written to manage images coming in with shows.
I needed to do more though, so this was not really good.
It only dealt with JPEG and it couldn't let me do some of the other things I needed to do.
A couple of the things that I do is to interrogate the image, to find its size.
And that's a thing that's not visible unless you run some sort of an analysis.
It's tool against it.
And if it was bigger than a certain size, I want to generate a thumbnail.
Because when I do this for people's images, and the image is too big,
I place a thumbnail in the notes and then that links to the full size image
which you can get if you click on the thumbnail.
So I've been using graphics magic, the package, which has got tons and tons and tons of features
for manipulating images.
It's a fork of image magic, image magic, I think it's still alive, but it's very buggy.
So they've done a lot of work to make graphics magic a lot more efficient and less, less buggy.
So this package can deal with the orientation transformation that I needed
and it can handle many other image types, not just JPEG.
So I discovered that the technique I needed to use was to use the GM is the command.
But you follow that with an action.
So GM space convert will take an image and convert it into another image.
There are options, hyphen strip, which removes all profiles and text attributes from the image.
So it takes X if out and any other things that you might find in there.
And the other option is auto hyphen orient, which rotates the image by actually transforming the pixels.
So it's upright and it adjusts the image orientation setting so that it's suitable.
Well actually it takes out the orientation setting, but before it strips everything out,
it does the physical transformation of the image so that it ends up being the right way around.
So I've got an example here, GM space convert hyphen strip hyphen auto hyphen orient.
And these are just dummy files.
Sideways underscore pick.jpg is converted to normal underscore pick.jpg.
So running this on the images that I was trying to resolve the state of did what I needed.
It removed the X if orientation after having rotated the pixels to the horizontal of the normal state.
So conclusion then, I've modified my picture management script to use this technique.
So far it seems to do the job perfectly.
Though I have to admit that images with X if orientation metadata are rare.
I've just been doing it against running it against test data so far.
The graphics magic documentation does indicate that transforming pictures from one orientation to another
can cause problems with some images.
I think it means if you've got something like a panorama type thing which is much, much wider than it is high,
then rotating it will do some very strange things to it potentially.
But I don't fully understand what the details are there.
If you know more then that would be interesting to hear it.
But as far as my script absurd, I think it's the problem is pretty much solved.
The other thing to say is I gained several things from the journey of discovery that this took me maybe a week-tender
in between other things to get my head around this completely.
So I ended up understanding images and how they work a bit better.
I ended up using X if tool more than I'd ever used before.
There is a way that you can call it to just do the full stripping of stuff from images.
I also discovered that if I opened the example image with GIMP,
it noticed that it had an X if orientation of 6 and it showed both the rotated version
and the original version of the image and asked if I wanted to perform the transformation
and take out the rotation setting.
Which I thought was amazing.
I wish I had seen that to start with and I would have understood it a bit better.
I also discovered a KDE image tool called Show Photo.
It's about with an F and it also reported that the image existed in two forms.
And in this case it allowed me to do editing of X if stuff through a GUI.
So that's quite nice. X if tool can do all of this.
But it's a Perl script actually and it's got a bazillion options to it.
So I find it a little bit difficult to fully understand.
Or even to partially understand.
Your mileage may vary.
So it was useful exercise.
Hopefully it's giving you a little bit of insight and maybe it's a thing
that might help you to solve a problem in the future.
I hope so.
Anyway, that's it.
Bye now.
You have been listening to Hecker Public Radio
at Hecker Public Radio.
Today's show was contributed by a HBR listener like yourself.
If you ever thought of recording a podcast,
then click on our contribute link to find out how easy it really is.
Hosting for HBR has been kindly provided by an onsthost.com,
the Internet Archive and our Sync.net.
On this otherwise stated, today's show is released under Creative Commons
Attribution 4.0 International License.