Report from the Third International Workshop on Computer Vision Meets Databases (CVDB 2007)

Laurent Amsaleg, Björn pór Jónsson, Vincent Oria
2008 SIGMOD record  
The goal of the CVDB workshop series is to foster interdisciplinary work between the areas of computer vision and databases. We have observed that few researchers in the computer vision community are adopting any of the indexing schemes designed by database researchers. Furthermore, while new and exciting techniques are being developed by computer vision researchers, database researchers are often unaware of such work. The reason is that, unfortunately, there has been a great gap between the
more » ... puter vision and database communities. The goal of the CVDB workshop series is to bridge this gap. The idea is to provide database researchers with a snapshot of what computer vision people are dealing with and vice-versa, with the aim of defining research directions that can benefit both communities. There is great expertise on both sides, and the CVDB 2007 workshop was aimed at sharing it by means of keynote speeches, technical presentations, and panel discussions. Workshop Program We assembled an international program committee of 27 experts from the computer vision and database communities. As was reportedly the case with many other workshops co-located with SIGMOD/PODS 2007, fewer papers were submitted than in previous years. Thus the program committee had to review only nine submitted papers, and in the end, four papers were selected for presentation and publication. Additionally, we hand-picked two keynote speakers to present their views of the research directions and contri-butions of the computer vision and database communities. Finally, we assembled a panel to focus on the current and future roles of content-based multimedia retrieval. After a short introduction, the day started with the first keynote speech on large-scale multimedia retrieval, followed by a technical session with the four papers. After lunch, the second keynote speech, on modeling events with media evidences, was followed by panel discussions. For details of the papers, tutorials, and panel, please visit the workshop web-site, which will remain open at cvdb07.irisa.fr. The CVDB 2007 proceedings appear in the ACM Digital Library. In the following, however, a summary of the main points of the workshop is presented. Keynote I: Large-Scale Retrieval The first keynote speech, titled "Challenges of and Remedies for Large-scale Multimedia Information Retrieval" was delivered by Edward Chang, director of research at Google, Beijing. According to Edward Chang, with the rapid growth of image and video data, it is increasingly crucial to provide tools that can assist with effective organization and search. Despite advances in several areas, challenges remain for the deployment of a Web-scale multimedia search engine. His presentation described three major challenges and their potential remedies. The first challenge is image and video annotation, which he claimed is necessary since most users prefer keyword-based search over content-based search. Manual annotation can be subjective and error-prone, whereas machine annotation cannot effectively discover all the desired information. Recent efforts, such as the ESP game, have moved towards fusing human and computer intelligence for improved annotation accuracy. The second challenge is that measuring similarity, in particular perceptual similarity, is difficult in many cases. For instance, image features can vary based on size and quality of the images. Work on feature constancy can potentially remedy this challenge. The third challenge that hinders the deployment of a large-scale system is scalability itself. A multimedia 46
doi:10.1145/1374780.1374793 fatcat:kqu4ek2smvhv3cjfqdwsfsywha