QuickTopic free message boards logo
Skip to Messages


Angular Radial Partitioning / Line Face Map

09:11 PM ET (US)
uggs sale ugg boots ugg australia ugg boots sale
  Spam messages 10-8 deleted by QuickTopic between 08-18-2010 02:03 AM and 05-17-2008 10:16 AM
12:23 AM ET (US)
Thanks for your discussion! BTW, this is very interesting sites: colchicine lawyer webpage devoted to colchicine lawyer. smoking opium webpage devoted to smoking opium. ... You may find it useful!
Deleted by topic administrator 07-21-2006 09:02 AM
Ben Laxton
04:56 PM ET (US)
IN the radial partitioning paper they present results of image retrieval that initially look good, but the recalled images are just rotated versions of the input. With these results they have shown that their method is a rotation invariant representation, but how well does it generalize - i.e. find a similar object in another context?
Erik Murphy-Chutorian
04:45 PM ET (US)
The Edge image description paper claims that the system is "Scale-Invariant". On page two, it says that scale invariance is achieved by "normalizing the edge map to WxW pixels". In other words, if I present the same image at two resolutions, the system can normalize them to the same size. This is hardly scale-invariance. In fact, this only could work if you surmise that the object has been perfectly segmented (or at least provided with a cropping rectangle), but the authors claim that their system avoids "segmentation preprocessing" , but unless I am not understanding it correctly, it appears that every object IS provided with a segmentation by bounding box. There is no reference to a multi-scale global search e.g. Viola and Jones either.
Reading this, it appears that the authors are not up to date with the literature, and that many relevant references are missing.

This is in great contrast to the PAMI paper which is well-written and insightful.
Edited 10-27-2005 04:51 PM
Brendan Morris
02:28 PM ET (US)
I'm not very impressed by the results because it shouldn't be too difficult to recognize the same image in slightly different positions. ARP is claimed to be robust to rotation, translation, and erosion yet they show in fig 4 the case where this doesn't hold. They don't show why q52 is a good query and what made q10 a poor search. I'm not very confident in the ablility of this method to actually do more complicated queries, aka a query with a different image than on from the database. Yet the retrieval results seem quite good. As for the speed constraint it does't seem to mean much for the video search criteria they reference. Based on other papers we've seen SIFT works quite well and this would probably be put in to some text retrieval framework so computation time isn't that important except for the 1 sketch query image. What I would be interested in seeing is a little more discussion about the binning/partioning with respect to orientation and maybe more importantly object size.
Robin Hewitt
10:19 AM ET (US)
In the LEM paper, it looks like they're using only lines at the highest resolution. It might be interesting to also match lines at other scales.

The ARP method is interesting, but it seems very special-purpose. Since it's an FT approach, it needs to combine information over a large area. I imagine you'd need to have very good centerpoint localization and limited changes in aspect ratio. Aspect ratio for face regions does change with expression, I've found, and I'd think the centerpoint could shift as well.
Anup Doshi
03:23 AM ET (US)
For the angular radial partitioning paper: wouldn't using SIFT and others already solve this problem? Although, it seems that this method performs very fast. It would be nice to see this work on non-contrived data.

For the Face Recognition: it seems to me that this LEM is inherently only using shape/structure as features. While it performs extremely well (relatively), would it help even more to include other features as well?

Print | RSS Views: 2515 (Unique: 1153 ) / Subscribers: 0 | What's this?