I've put up some classes for storing content based MPEG-7 image
descriptors in a lucene index and querying the stored descriptors to get
"similar" images. In other words: I've put up a simple library for
content based image retrieval powered by lucene.
The performance tests are quite promising as the retrieval is based on a
linear search, but I'm afraid it is not suitable for enterprise
searching. In my opinion searches in ~ 100.000 images should be ok.
I'd appreciate feedback (as I just put it in 2 evenings :-) on:
* How useful is the API?
* Does it work well for use cases you imagine?
* Is it fast enough?
* Are the returned results useful?
Known issues: One of the three MPEG-7 descriptors (namely EdgeHistogram)
is painfully slow. I also identified the main bottlenecks, but I didn't
get around yet fixing it.