Abstract
In many contexts, it is desirable for users to be able to extract content directly from image and video data via queries formed dynamically online. Queries over pre-computed features of an image set restrict the user's access to what has been computed by an editor or supervisor of the data. This paper presents an approach to extracting patterns from images and video efficiently in order to support online queries directly on the original image data. Our technique relies on compressed-domain processing, where we perform content extraction on image or video data that are compressed with standard algorithms such as JPEG, MPEG or wavelet schemes.
Original language | English |
---|---|
Title of host publication | Proceedings - IEEE Workshop on Content-Based Access of Image and Video Libraries, CBAIVL 1997 |
Pages | 52-58 |
Number of pages | 7 |
ISBN (Electronic) | 0818679832, 9780818679834 |
DOIs | |
State | Published - 1997 |
Event | 1997 IEEE Workshop on Content-Based Access of Image and Video Libraries, CBAIVL 1997 - San Juan, United States Duration: Jun 20 1997 → Jun 20 1997 |
Publication series
Name | Proceedings - IEEE Workshop on Content-Based Access of Image and Video Libraries, CBAIVL 1997 |
---|
Conference
Conference | 1997 IEEE Workshop on Content-Based Access of Image and Video Libraries, CBAIVL 1997 |
---|---|
Country/Territory | United States |
City | San Juan |
Period | 6/20/97 → 6/20/97 |
Bibliographical note
Funding Information:In this paper we have shown that significant speedups are possible in solving the problem of matching patterns in compressed image data. The speedups are available by operating on the data while it is compressed. Without errors from quantization, which is involved in most lossy-coding schemes, the classification results are exact. But even with significant levels of quantization, the eigenspace which serves as the classifier remains robust and performs well. When the eigenspace projection dominates the computation time, as is the case when a small template is matched at all locations in a large image, we achieve a speedup of a factor of 2. In other experiments where a small number of image templates from each compressed image are projected to eigenspace, the decompression time plays a significant role as well and leads to speedups on the order of a factor of five. Acknowledgments We gratefully acknowledge the support of the National Science Foundation for this research under grants IRI-9308415, CDA-9320179 and CDA-9502645.
Funding Information:
We gratefully acknowledge the support of the National Science Foundation for this research under grants IRI-9308415, CDA-9320179 and CDA-9502645.
Publisher Copyright:
© 1997 IEEE.
ASJC Scopus subject areas
- Library and Information Sciences
- Information Systems
- Signal Processing