ARDA VACE Video Information Summarization and Demonstration Testbed Howard D. Wactlar (PI) Mike Christel, Alex Hauptmann, Jianbo Shi Carnegie Mellon University, USA
VACE Goals
Informedia Overview
PPT Slide
System Screen Shot
Data Collection: The Vest
Application of Diverse Technologies
Challenge: Imperfect Technology
Informedia VACE System Overview
Combined Technologies Integration
Multiple Levels of Abstraction & Summarization
Speech Recognition Functions
Speech Recognition Accuracy
Speech Variations
Information Retrieval Recall vs. Speech Recognition Accuracy
Information Retrieval Precision vs. Speech Accuracy
Information Retrieval
KNN-based Topic Detection
Image Understanding Functions
Video OCR
Video OCR Block Diagram
VOCR Frames
VOCR Preprocessing Problems
VOCR
Video OCR - Results
“Name-It” Face/Name Association
Image Search & Similarity Matching
Similarity Challenge
Images Containing Similar Colors
Images Containing Similar Shapes
Images Containing Similar Content
Goal: Automatic Video Characterization
Annotations
Informedia Current Capabilities
New VACE Directions
VACE Initiatives
Integration Platform
Distributed Infrastructure
Informedia System Structure
(Current) Informedia Processing Flow
Informedia Data Hierarchy
Informedia DB Schema
Common Data Format
Display-1 Example: XSLT-Produced Image Grid Display
Display-2 Example: XSLT-Produced Large Image with Text Display
Challenge of Video Information Summarization
Summarization Subgoals
Metadata for Video Summarization
Named Entity Extraction
Summarization and Visualization Dimensions
Multimodal Search: Word Search and Map Search
Query-based Poster Frame Selection Process
“Skim Video”: Extraction of Significant Content
Visualization of Result Space
List of Active Topics
Visualization Features
Map Digest View
Map Collage
Timeline Collage
Link Analysis
VACE – Year 1
VACE – Year 2
Email: informedia@cs.cmu.edu
Home Page: http://www.informedia.cs.cmu.edu
Other information: Please send us email if you would like to request higher resolution slides.