Lire 0.7 is a major release fixing a lot of bugs and introducing several new features including new descriptor, a simplified way to use descriptors by introducing new generic searchers and indexers as well as an generalized interface for image descriptors. There are also several improvements in indexing and search speed (especially in autocolorcorrelogram). Furthermore retrieval performance was optimized based on the Wang 1000 data set. If you use Lire 0.7 to update an existing version, please make sure that your indices are created newly from scratch. All new features have also found their way into LireDemo, which now also supports multi-threaded indexing.
Thanks for comments, code and general comments go again to Savvas Chatzichristofis, Rodrigo Carvalho Rezende, Marko Keuschnig, Christian Penz, Oge Marques, Anna-Maria Pasterk and Christoph Kofler.
Recently Savvas Chatzichristofis released a Facebook app based on his tool img(Rummager) to retrieve the photos of Facebook contacts based on image content. The tool – called img(Finder) – allows to connect to a Facebook account and creates an index based on the CEDD and FCTH image features. img(Finder) indexes the images of all Facebook friends. Querying the index is done by sketch (painting tool is included) or by sample image. Best of all: The tool is open source 🙂
Since Blobworld was down I was hoping for a new online CBIR system to show in lectures. Last week I received word about the img(Anaktisi) image search engine from Savvas Chatzichristofis. This search engine is based on 2 new content based descriptors that combine color and edge features. The results look promising!
There is also an offline sketch based retrieval system being developed. A screencast can be found on YouTube.
Last Monday I’ve been at the Text Information Retrieval workshop, taking place in the Regensburg University in context of the DEXA conference. Benn Stein – as always – did a very good job on organizing the workshop. There were 11 papers accepted and the overall quality was high. All paper can be found online here.
Personally I’d recommend the following if you want to take a look at the highlights of the workshop:
Furthermore I also presented a paper there: Aspects of Broad Folksonomies, which I’ve done together in constant collaboration with Michael Granitzer and Roman Kern. The slides are also online here.
On 3rd September the Workshop on Text Information Retrieval (TIR 07) takes place in Regensburg, Germany. I’ll be there presenting our work (Michael Granitzer, Roman Kern and me) on folksonomy characteristics: Aspects of Broad Folksonomies (download pdf). Main contribution of the paper is that tag similarity in the del.icio.us folksonomy follows a power law in many cases. This means that many tags have only few highly related tags, while the similarity to other tags is rather low.
You can find the whole program as well as the contributions (as pdf) on the TIR07 website.
In the last issue of the SIGIR Forum I found a very interesting in my opinion is the article of Massimo Melucci: On Rank Correlation in Information Retrieval Evaluation. I really like this paper. Massimo Melucci gives a lot of basic references on the use of rank correlation in retrieval evaluation. However his publication has a very practical (in contrast to mathematical & theoretical) essence and remains restricted to the comparison of values generated by experts or algorithms. The complicated task of integrating the target user (not an retrieval expert assessing the results) is not discussed in length.
Especially the last thing was part of the work in my PhD. One can find some further information in the publication Empirical Studies in Multimedia Retrieval Evaluation available here.
As Frank Smadja pointed out here (see comment) the original proceedings of the tagging workshop @ WWW 2006 site is down. Luckily the files are not lost, but mirrored here:
In Lire 0.5.4 some bugs were fixed: The scalable color descriptor (color histogram) was not compliant to the MPEG-7 standard, which is now fixed. The color only search was changed to use the color layout descriptor and a bug in the edge histogram descriptor was hunted down.
Note that you have to re-index your files: Your old index cannot be used with the new version as 2 descriptors have changed. Furthermore all binary files have been compiled with Java 6.0. So if you need a Java 1.5 version you’ll need to recompile yourself (ant build file is included) and include the swing layout class library from NetBeans.
The LireDemo GUI application has also been updated: A new function for creating image mosaics has been introduced and the indexing of digital photos is now faster than ever as only the EXIF thumbnails – if available – are used instead of the whole image.
Go to the LIRe page for download links and further information.
As I really like the field of information retrieval, I ordered a new book for our University library: Google’s Pagerank and Beyond: The Science of Search Engine Rankings (Amy N. Langville & Carl D. Meyer, University Presses of CA). Wondering if this was a good one I took a look inside …
I’ve got to say: I’m impressed! All the main ideas of information retrieval and PageRank (as well as HITS) are described within the first few chapters. The following chapters go into mathematical and technical details: Why and how does this work, why and how can it be used in a distributed environment? My opinion: Two thumbs up!
At the WWW 2006 a workshop on the topic tagging took place. There were several papers especially on folksonomies and emergent semantics.
The proceedings are available online at: /hosted/taggingws-www2006-files/taggingworkshopschedule.htm The proceedings are hosted here as the original site is no longer available. (Many thanks to Frank Smadja, who provided the files and allowed to mirror them)