Nautilus Thumbnails and CBIR

16 August, 2004 at 06:55 Leave a comment

Just following-up to your thread on GNOME mailing lists about providing
thumbnail support for Nautilus –
Actually, I think it is a cool idea. Because I am working on personal
information retrieval and dudring the course of an experiment I found
out that doing content based image retrieval on thumbnails actually does
prove useful to find files which “look similar” and so the files could
be of similar content as well.
For example, say I have some white-papers from some company like
Factiva. They may be downloaded years apart but the policy of Factiva
forces the whitepapers to have a similar look. So, getting all files
from Factiva (and say, whitepapers), is to just search for images which
look similar. Of course, there will be false hits but we can sort them
out. I have 2 years to figure that out πŸ˜‰

* I am actually looking for a small utility sort of
thing which takes a directory as input and gives me the thumbnails of
all the files as output with simple filenames for tags. I will then
index them using simple/complex CBIR techniques for future searches

1) Do you know of any such utility to do that? I have seen plenty of
utilities which take a directory of JPEG images and create thumbnails
and web-pages but I want all (supported) files – Office, PDF, HTML… to
have thumbnails generated

2) Can your tool be extended to do that? Can you please lend me the
source and how to go about it? I do not know much coding (I am more of a
mathematician) and so your help would be much appreciated

3) I was also thinking of a generic algorithm which takes a directory,
opens the file by its default handler in full-screen mode if possible,
snatch a screenshot and then store the thumbnailed version. Ideally this
should work in Java. What do you think? Don’t you think this works much
more generic than most things? We can have OS specific handlers and
there is a Java program I think I saw sometime before which can grab
screenshots. Of course, if you wish you can write a fast script. Please
do let me know if you intend doing so

4) Interestingly, kio_thumbnails does a pretty good job (no proper
Office support though). It stores all thumbnails in
~/.kde/share/thumbnails folder in some arbitary naming scheme. There are
two problems here. There is no way to know which images belong to which
directories and there is no command like kio_thumbnail (although and exist in /usr/lib/kde3). There
should be some way to make a binary from this I suppose. Any idea where
I can look?

4) If you are not convinced of my argument, please do install imgSeek –
and index the KDE thumbnails directory and play around with the image
search feature. I am sure that will make things clear πŸ™‚

I think that GNOME, Linux and OSes should look beyond
normal ways of text and metadata searching. That is my thesis anyway.
For example, you should have a context-sensitive menu which gives you
options like “search for files which look similar”. Note that KDE
already has a “search for similar images” (which I could never get to
work). We could also have something like “search for similar sounds”
(there are some things out there which do this I suppose) but what we
could do is basically dump out some information (like say the
“signature” of audio files) doing some fairly simple processing and then
construct an index which can search that. I think “signatures” can be
seen as a generic framework. For images, thumbnails are signatures. For
audio files, something else could be. I hope you guys get the point and
try to help me here. And also spread the meme so that future versions of
GNOME might have this kind of searching built-in…

Phew! that was a long post. Sorry about that. I look forward to your

Srikant (


Entry filed under: Computers/ICT, Glasgow-Travails, Projects, Research, WebXP.

League of MBA Bloggers Sony’s Walkman Impact

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

Trackback this post  |  Subscribe to the comments via RSS Feed


August 2004
« Jul   Sep »


%d bloggers like this: