ACM SIGIR'98 Post-Conference Workshop on

Multimedia Indexing and Retrieval

Melbourne, Australia, August 28, 1998

Call For Participation

Background

About the workshop

The first focus of this workshop is on integrating information from various media sources in order to handle multimodal queries on large, diverse databases. An example of such a collection would be the WWW. In such cases, a query may be decomposed into a set of media queries, each involving a different indexing scheme. The interaction of various media sources that occur in the same context (e.g., text accompanying pictures, audio accompanying video) is of special interest; such interaction can be exploited in both the content analysis and retrieval phases.

The second focus deals with examples of research using content and organization of multimedia information into semantic classes. Users pose and expect a retrieval to provide answers to semantic questions. In practice this is difficult to achieve. Building structures that encode semantic information in a fairly domain independent and robust manner is extremely difficult. A quick review of computer vision research over the last few years points to this difficulty. In many cases, image content can be used in conjunction with user interaction and domain specificity to retrieve semantically meaningful information. However, it is clear that retrieval by similarity of visual attributes when used arbitrarily cannot provide semantically meaningful information. For example, a search for a red flower by color red on a very heterogeneous database cannot be expected to yeild meaningful results. On the other hand retrieval of red flowers in a database of flowers can be achieved using color. In context therefore, examples of research using content and organization of multimedia information into semantic classes will be discussed.

Many systems, particularly image and video based ones require an example picture which can be used as a query (alternatively, the user may be required to draw a picture). It may be unrealistic to expect an example image to be always available. Thus, it would be useful to find ways of generating new queries. Can NLP techniques be combined with computer vision techniques to generate such queries? Or can multimodal retrieval techniques be combined to create queries suitable for image, video and audio retrieval? In general, a question is how can we create realistic queries for realistic systems.

The third focus of this workshop is on evaluation techniques for multimedia retrieval. Currently, most researchers are using the standard evaluation measures defined for text documents; these need to be extended/modified for multimedia documents. There is also a high degree of subjectivity involved that needs to be addressed.

We will focus on the following specific topics:

• content analysis and retrieval from various media (text, images, video, audio)
• interaction of modalities (e.g. text, images) in indexing, retrieval
• effective user interfaces (permitting query refinement etc.)
• evaluation methodologies for multimedia information. We have found that researchers pay insufficient attention to it.
• techniques for relevance ranking
• multimodal query formation/decomposition
• logic formalisms for multimodal queries
• indexing and retrieval from scanned documents - e.g extracting text from images, word spotting - as a retrieval technique for both handwritten and printed documents.
• testbeds for evaluating multimodal retrieval: it would be nice to have some resource sharing here since annotating these, and coming up with a good query set are difficult

Participation

Two types of participation are expected. Those interested in making a presentation at this workshop should submit their full papers either in online postscript version or in hardcopy by regular mail to the address given below. The papers should not exceed 5,000 words, including figures, tables, and references. Those interested in participating, but not presenting papers, should submit a statement of interest, not to exceed 500 words. This should clearly state what aspect(s) of the workshop reflect their research interest. These will be used to select panelists. Both types of submissions are due on Friday, June 5th. Decisions will be made no later than Friday, June 26th. In the case of paper submission, the final camera-ready papers are due on July 24th. Working notes will be made available to all participants at the workshop. All the submissions should be sent to:

Prof. Rohini K. Srihari
CEDAR/SUNY at Buffalo
UB Commons
520 Lee Entrance, Suite 202
Amherst, NY 14228 - 2583
Email: rohini@cedar.buffalo.edu
Phone: (716) 645 6164 ext. 102 Fax: (716) 645 6176

Organisation

Workshop chairs (also program chairs):

Rohini K. Srihari	CEDAR, SUNY at Buffalo
	Amherst, NY 14228 - 2583 rohini@cedar.buffalo.edu
Zhongfei Zhang	CEDAR, SUNY at Buffalo
	Amherst, NY 14228 - 2583 zhongfei@cedar.buffalo.edu
R. Manmatha	Computer Science Dept., Univ. of Massachusetts
	Amherst, MA 01003 manmatha@cs.umass.edu
S. Ravela	Computer Science Dept., Univ. of Massachusetts
	Amherst, MA 01003 ravela@cs.umass.edu

Program Committee Members:

Shih-Fu Chang	Columbia U., USA
David Harper	Robert Gordon University, U. K.
Alex Hauptmann	CMU, USA
Rakesh Kumar	Sarnoff, USA
Desai Narasimhalu	ISI, Singapore
Candace Sidner	Lotus, USA
Peter Schauble	ETH, Switzerland

Timetable

Further information

Further questions may be directed to the address above, or go to the Web page of this workshop , or the SIGIR Conference main Web Page .