Peekaboo

In  Szeliski book I find more and more interesting references. For example a paper by Leyvand, Cohen-Or  and Dror that describes how to warp a portrait into a "more pleasing" portrait by moving features according to images that were rated as being beautiful. And in my subjective impression it is working quite well. The modifications are quite subtil but make a big difference on the overall impression. I don't really want to think about the implications of the existance of such an algorithm... but it is quite interesting anyhow.

After reading the introduction to Szeliski excellent book , I started browsing through visionbib , which he recommends as a good resource. It is quite a big library but seem a little inaccessible to me. There is are databases of available implementations, papers and datasets and each is quite expansive. But even though there are RSS feeds I still have the impression someone missed out on the web 2.0 (or maybe even 1.1).

There are some great videolectures online from CVPR this year. Two I find particularly intresting are Layered Object Detection for Multiclass Segmentation and Cascade Object Detection with Deformable Part Models . The former basically uses sliding window classifiers for detection and shape priors to generate segmentations from detections. The second is an interesting continuation on Felzenszwalb's part based models . The talk explains how to use a cascade to speed up detection.

On the connectionist mailing list Geoffrey Hinton , the father of the Restricted Boltzmann Machine, posted a guide on how to use and train them . Quite nice if you are interested in these kinds of models. But keep in mind: Restricted Boltzmann Machines are Hard to Approximately Evaluate or Simulate. No matter what the guys in Torronto say ;)

Today I stumbled upon a book on Computer Vision that Richard Szeliski from Microsoft Research is publishing this fall. There is a current draft available on the website and it looks great. It goes through all the basic concepts and is full of examples. Definitely worth checking out if you don't know about it yet.