SemanticMetadata.net

How users think about image search …

There is an article about some new image classification and labeling system online: The UC San Diego has developed (together with Google, they brought in the data) a machine learning approach for automatic image labeling. Find the article here.

Well this is not the first image analysis engine that was created and it won’t be the last, but the discussion at slashdot was particulary funny. Some excerpts go here:

• I remember when we had to go to a gas station and *buy* porn. Now you have computers out there finding porn for you. You kids today have it too easy!
• If this doesn’t revolutionize the searching of online porn galleries, I don’t know what will.

• … was similarly trained to recognize tanks in landscapes. [...] Then they introduced it to a new batch of images and it fell apart. Turns out that the initial set of images had all the tanks shot on a sunny day and all the tankless images shot on a cloudy day (or vice versa). It had learned to tell a sunny day from a cloudy day. Ha ha.
  

• Now I can search for porn stars that look like that girl in my English class!

Thanks to social software we now know how the ordinary talkative geek thinks about image search

(thx to Roman for the hint on the article & the discussion)

CfP 4th International Workshop on Text-Based Information Retrieval

Like in previous years Benno Stein organizes the TIR, the Workshop for Text Information Retrieval. This year the workshop takes place in conjunction with the 18th International Conference on Database and Expert Systems Applications (DEXA 2007) in Germany in September.

Call for Papers (see also here):

Intelligent technologies for information mining and retrieval have become an important and exciting field of research in our information-flooded society. Methods of text-based information retrieval receive special attention, which results from the fundamental role of written text, from the high availability of the Internet, and from the rising importance of the different forms of Web communities.
Continue reading CfP 4th International Workshop on Text-Based Information Retrieval…

Report: Context & Digital Photo Collections (MMC-Workshop in Aachen)

Today the Workshop on “Multimedia Metadata – The Role of Semantics” started in Aachen, Germany. Currently the keynote presentation of Susanne Boll from Oldenburg University takes place. Susanne speaks about the necessity of taking the context of digital photos into account while analyzing photos and extracting metadata. She also gave examples: While indoor / outdoor classification is quite a challenging task, the combination of EXIF information (context dealing with flash, shutter speed, shooting mode, focal length, etc. ) with the content analysis yields quite good results.

The talk of Susanne shows the direction where multimedia research should (also) go. The integration of the user is necessary as it is the driving force of the quite successful Web 2.0 movement.

Lire 0.5.2 Released: Auto Color Correlogram

The 0.5.2 release of LIRe brings along a new descriptor, which is kind or “more advanced version of a color histogram”. The  so called color correlogram is based on the probability to find pixels of certain colors in certain neighborhoods. Leaving the theoretical part aside the color correlogram is a new way to retrieve photos with LIRe based on color and color distribution, which might be very interesting for applications heavily depending on colors. Further information on the correlogram might be found at the development Wiki.

Lucene for Java 2.1.0 Released

The release of Lucene 2.1.0 has been announced recently. The new release includes bugfixes, performance improvements, new features and removed some deprecated things. You can find the whole list of changes in the CHANGES.txt file. Some new features of interest are:

• A new “Match All Documents” query option in QueryParser
• New methods fro handling updates in IndexWriter
• Support for leading wildcards in QueryParser

You can find the new release at lucene.apache.org/java.

Application of Interest: Retrievr

Content based image search in Flickr? Well someone (to be specific System One) already had the same idea. Based on a scientific project from 1995 Flickr images are indexed using signatures generated from a wavelet transform.

The resulting application is named Retrievr and can be used to draw query images, which are then used to find matching Flickr images. The system works quite smooth, so  give it a try! A description has also been published at slashdot.

Blobworld Testimonial

One of the most famous services for image segmentation & retrieval is (or was) blobworld. The system was built for advanced image querying based on regions (blobs), which were identified automatically. For image retrieval one could select blobs and define whether color, texture or size and position were most relevant for the query. Based on the input several good matches were presented.

Over the years Blobworld became a de facto standard for showing the possibilities of content based image retrieval systems. Hopefully there will be more innovative applications like this one in the future.

Call for Papers: Workshop on Text-Based Information Retrieval 2007 (TIR 07)

The CfP for the TIR 07, the Workshop on Text-Based Information Retrieval 2007, has been published. This year it takes place in Regensburg in conjunction with the Dexa 2007. You can find the full CfP at http://www.aisearch.de/tir-07/.

I also visited the last two TIR workshops and I’m looking forward to this years TIR 07. You can find links to the proceedings of the TIR 06 here. The proceedings of the TIR 05 are here.

Call for Papers: Workshop on Text-Based Information Retrieval 2007

Intelligent technologies for information mining and retrieval have become an important and exciting field of research in our information-flooded society. Methods of text-based information retrieval receive special attention, which results from the fundamental role of written text, from the high availability of the Internet, and from the rising importance of the different forms of Web communities.

Various techniques and methods are being used for text-based information retrieval tasks, which stem from different research areas: machine learning, computer linguistics and psychology, user interaction and modeling, information visualization, Web engineering, or distributed systems. The development of powerful retrieval tools requires the combination of these developments, and in this sense the workshop shall provide a platform that spans different views and approaches.

The following list gives examples from classic and ongoing topics from the field of text-based information retrieval for which contributions are welcome (but not restricted to):

• formal models for text representation, document models, similarity measures for special retrieval tasks
• category formation and clustering, document classification
• IR and natural language processing: topic identification, text summarization, keyword extraction
• Web community mining, social network analysis, collaborative tagging and IR
• plagiarism analysis, author identification, style analysis
• concepts and techniques for information visualization, user modeling, and interaction for particular retrieval tasks
• relevance feedback and personalization
• evaluation, building of test collections, experimental design and user studies
• multilingual issues in IR: cross-language retrieval, multilingual retrieval, machine translation for IR
• IR for the Semantic Web: usage, extraction, and maintenance of knowledge
• IR and software engineering: frameworks, architectures, distributed IR
• IR in business and engineering applications

The workshop addresses researchers, users, and practitioners from different fields: data mining and machine learning, document and knowledge management, semantic technologies, computer linguistics, and information retrieval in general. In particular, we encourage potential participants to present research prototypes and tools of their ideas.

see also http://www.aisearch.de/tir-07/

Color Based Image in E-Commerce Example

George L. sent me a nice example of a website integrating image search features: Become.com offers a search ability for color based retrieval of clothing. Just use the “find by color” feature in the right hand side column at the top and you will be impressed how much fun image retrieval can be .

One can also observe how inaccurate algorithms get when the background is integrated in the color histogram (e.g. with this search parameters, where I searched for grey and found also some blue images – or even in the presented screenshot where an obviously not red dress is shown before a red background). However this is a still unsolved problem: How to separate background and foreground in still images?

TIR-06 Proceedings in CEUR-WS available

As I’ve presented a paper on Caliph & Emir at the 3rd Workshop on Text-based Information Retrieval in Riva del Garda, Italy, Aug 29, 2006, I’d like to announce the publication of the proceedings on CEUR-WS.org. You will find all papers from the workshop on http://ceur-ws.org/Vol-205:

• A Framework for the study of Evolved Term-Weighting Schemes in IR, Ronan Cummins, Colm O’Riordan
• Integrating tf-idf Weighting with Fuzzy View-Based Search, Markus Holi, Eero Hyvönen, Petri Lindgren
• Syntax versus Semantics: Analysis of Enriched Vector Space Models, Benno Stein, Sven Meyer zu Eissen, Martin Potthast
• Graph Retrieval with the Suffix Tree Model, Mathias Lux, Sven Meyer zu Eissen, Michael Granitzer
• Classifying Encounter Notes in the Primary Care Patient Record, Thomas Brox Røst, Øystein Nytrø, Anders Grimsmo
• LexiRes: A Tool for Exploring and Restructuring EuroWordNet for IR, Ernesto William De Luca, Andreas Nürnberger
• Framework for Semi Automatically Generating Topic Maps, Lóránd Kásler, Zsolt Venczel, Lászlá Zsolt Varga
• Ensemble-based Author Identification Using Character N-gram, Efstathios Stamatatos
• Common Criteria for Genre Classification: Annotation and Granularity, Marina Santini
• Challenges in Extracting Terminology from Modern Greek Texts, Aristomenis Thanopoulos, Katia Kermanidis, Nikos Fakotakis