Video Retrieval Based on Language and Image Analysis
Abstract
This report summarizes the technical work ADVANCE Inc. performed under DARPA's contract no. MDA972-97-C-0014. This project has been conducted during an approximate two and one half year period in Phase I and Phase II implementation. The work performed has been focusing on the research and development of technologies for the indexing and retrieval of digital video content. The methodologies are built upon the theories of frame-based access and object-access, which allows users to index and retrieve video contents accordingly. Research is also performed on access at levels of video scene and video programs. For these purposes, algorithms for video shot boundary detection and keyframe extraction were developed. Closed captioning is extracted and associated with video shots so that keyword searching can be performed to retrieve video shots. Different approaches are used to segment video objects, including motion-based segmentation, and the combination of motion information and image cues for segmentation. Special objects such as human face and military target of interest were investigated as indexing points. Techniques for object identification and tracking were also studied for indexing and retrieving video contents. A comprehensive prototype has been built and software has been developed for frame-based access technologies.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 28, 1999
- Accession Number
- ADA364129
Entities
People
- Yiqing Liang