Re: Basic questions on indexing of images

man, 07,.03.2005 kl. 17.22 -0600, skrev Jon Trowbridge:
> As an enthusiastic user of f-spot, I would love to see better
> beagle/f-spot integration.  I don't know if a separate FSpotQueryable is
> the right way to go.  Since f-spot stores photos on the file system,
> they will already get indexed by the FileSystemQueryable.  Thus the
> FileSystemQueryable seems like the reasonable point of integration.
> At one point I did a crude hack to index f-spot data, but it is
> currently disabled.  (The code is still in beagled/FilteredIndexable.cs,
> but it is #if false/#endif-ed out.)  It seems to me that there are two
> reasonable ways to approach this problem, but both have problems:
> The first approach is the one I used: basically attaching the f-spot
> metadata to the file when it is indexed.  That let's you search the
> f-spot metadata for free, but there would have to be a way to keep the
> lucene index up-to-date if the f-spot db is changed.  If you can get
> fine-grained notification of f-spot db changes, this would probably be
> the way to go.  If not, it might be hard.

This is the approach I had in mind. Although, I don't quite understand
the problem. If we use a FSpotQueryable, we will have a worker which
watches the f-spot photo database for changes. When these occur, it will
update the info in the data stored by FileSystemQueryable. Since the
photo database stores both filenames and metadata, it would be possible
to get the indexed data from lucene based on the filename? If so, an
updater shouldn't be too hard to write.

> The second approach would be to do a separate search of the f-spot
> metadata at query-time, and then both adding the matching files to the
> list of hits that comes out of the lucene index and attaching the
> necessary metadata to the hits before the are passed on to the client.
> Some of the infrastructure to do stuff like this is already written, and
> should land in CVS soon.  But it also requires us to search the f-spot
> database ourselves, which could introduce other problems.  One option
> would be for f-spot to maintain a small lucene index for its tags, but
> Larry might not like that idea.  And getting notification of changes is
> still a problem.

Something like the google search? I thought about something like this
too, but isn't it possible we will get two hits on the same file? If I,
for example, have some photos called birthday[01-50].jpg and everyone
has been tagged with birthday in f-spot, wouldn't I get 100 hits were as
there are 50 duplicates?

Kristian Berg

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]