Capturing and organizing vast volumes of multimedia data, such as scientific and medical data, requires new information processing techniques in the context of pattern recognition and data mining. Despite the massive quantities of multimedia content available today, not enough progress have been made in building an systemized, operative, and accessible database of gathered data.
Our goal is to enable a multimedia system, i.e. an organized, easily accessible, and searchable database of large-scale amount of multimedia data. We offer some answers on how to efficiently access the data, how to perform similarity search in relevance feedback scenarios and how to summarize and characterize information content in images in order to discover underlying patterns. The methodologies we propose involve approaches whose origins lie in several disciplines beside classical data management, including optimization, information theory, pattern recognition, signal compression and image processing.
Large-scale, content-based image retrieval for the World Wide Web. The system includes several feature spaces, relevance-feedback and association rule mining. more...
» On-Line Demo v3 (Searching over 10,000,000 Web Images) - SOON!!!
» On-Line Demo v2 (Searching over 10,000,000 Web Images)
We propose a modified MPEG7 texture descriptor that has comparable performance, but with nearly half the dimensionality and less computational expense. We also propose a new normalization and bit allocation method that improves similarity retrieval and indexing efficiency. more...
We introduce the problem of repetitive nearest neighbor search in relevance feedback and propose an efficient search scheme for high dimensional feature spaces. We propose a search algorithm that supports relevance feedback for the general quadratic distance metric. more...
This work introduces a novel approach to spatial event representation and analysis for large image datasets. more...
Our objective is develop models to describe geo-spatial object classes (e.g. airports, golf courses, harbors, etc.) in order to enable tasks such as object detection, recognition, and segmentation. We created statistical models for various geo-spatial objects with image texture as the visual feature. more...
This project investigated multi-modal and multi-feature access in images collected from the WWW. Relevance feedback was used to iteratively improve the results online. more...
» Demo (include 607,000 images)
© 2000-2006 Vision Research Lab, UCSB
Last updated:
Credits & Contact
Page visits:
View Stats