Image indexing and retrieval: Current projects and a comprehensive research agenda for the future



This panel focuses on the major research questions needing further exploration in the areas of image organization, retrieval, and use. The panel will first have short presentations on several ongoing image research projects and presenters will briefly comment on their current research, new tools and approaches to image indexing, and the broader research areas they address. The panel will then move into an interactive mode and the moderator will present a brief outline of a broad-based image research agenda for panel/audience dialogue, through which the agenda will be expanded and refined. In particular, current research in image indexing and retrieval focuses on the conference topic of “Thriving on Diversity – Information Opportunities in a Pluralistic World” as many newer tools (e.g., geotagging) and many new voices are joining in the image description process. The brief presentations relate to topics now appearing in the image indexing literature: information and knowledge behavior in diverse contexts, social networking in a linguistically and culturally rich environment, and challenges of harmony versus hegemony, as well as quality and relevance to particular audiences.

A Model of Image Needs, Retrieval and Use Across Several Professions, Joan E. Beaudoin

This investigation examines several image user groups to develop a theoretical model to clarify users' image needs, retrieval and use. By examining several groups of users (archaeologists, architects, art historians and artists), who are comparable in their heavy reliance on images of cultural materials to perform their work, the research will clarify the similarities and differences among image users' behaviors. The information gleaned from this investigation will in turn help explain how extensive the influence of discipline and task are on users' image needs, retrieval and use. Beyond adding to the limited body of knowledge about information behaviors surrounding images, the research will expand our understanding of specific user needs to inform system design. This knowledge is critical since image systems need to be developed with a fuller understanding of users' needs, the tasks they perform and their frames of knowledge. With the proliferation of images witnessed in recent years comes an increased need for research to clarify and explain image users' needs, modes of retrieval and use of visual information. This study will begin work in this direction by developing a comprehensive framework of image users' behaviors.

Indexing and Retrieving Images in a Multilingual World, Elaine Ménard

This research project compares two different indexing approaches for ordinary images of everyday life objects, namely: traditional image indexing with the use of controlled vocabulary or free image indexing using uncontrolled vocabulary. The experiment seeks to evaluate the performance of each approach for image retrieval. The use of controlled or uncontrolled vocabularies raises a certain number of difficulties for the indexing process. These difficulties will necessarily have consequences at the retrieval stage. This research also compares image retrieval within two contexts: a monolingual context where the language of the query is the same as the indexing language, and a multilingual context where the language of the query is different from the indexing language. This research used three data collection methods: analysis of the indexing terms in order to examine the multiplicity of term types applied to images (generic description, identification, and interpretation) and the degree of indexing difficulty due to the subject and the nature of the image; simulation of the retrieval process with a subset of images indexed according to each indexing approach studied; and finally, submission of a questionnaire to gather information on searcher satisfaction during and after the retrieval process. The findings of the study reveal if one of the two indexing approaches surpasses the other, in terms of effectiveness, efficiency, and satisfaction of searchers trying to retrieve ordinary images in a multilingual context.

Geotagged photographs: Can we standardize our representations of the world?, Diane Neal

This presentation will feature the results of a study involving the analysis of geotags as they are used on the photograph sharing Web site Flickr. The process of geotagging involves describing and identifying the geographic place where a photograph was shot. Millions of pictures have been geotagged on Flickr, but no standards have been developed for describing the format, specificity, or other features of a photograph's geographic location. For example, some pictures are geotagged with latitude and longitude, or with altitude information. Other geotagging approaches include the utilization of user-assigned descriptors such as “Portugal,” “New York,” or “near 25th and Main.” These and other variations in geographic representation may lead to low recall in Flickr searches for geographic locations. Geotag creation methods also vary greatly. Informed by applicable theories such as basic level theory and faceted classification theory, a data analysis of representative geotagged pictures will precede a discussion of the most prevalent mental models of Flickr geotagging. Possible approaches to reconciling uncontrolled geotagging and authoritative methods of geographic description will be explored.

Measuring and Enhancing Value of User-Created Metadata, Besiki Stvilia and Corinne Jörgensen

Social content creation systems such as Flickr blur lines among the traditional roles of information creators, intermediaries and users. Users increasingly participate in information creation, organization and curation. This makes it possible to accomplish tasks that still may need a significant human involvement (e.g. describing/annotating biological species, or photos) at a relatively low cost. Users, however, need to be motivated to create not only more, but also higher quality metadata. Research is needed to understand what the current practices of and motivations for metadata creation and organization by users are, and where and how one could intervene to improve this metadata. Similarly, studying user metadata/tagging practices might help harmonize and better align standard metadata and knowledge organization tools with user needs. Finally, for effective resource allocation, there is a need for defining systematic models and metrics for measuring the value of metadata and metadata quality. The researchers will discuss findings from their recent studies on photo tagging and collection building practices in Flickr, and how these practices may inform design and upkeep of standard metadata schemas and vocabularies. They will also present a model and metrics for assessing the value of metadata and metadata quality.


Sponsored by SIGVIS