Real-time hypothesis driven feature extraction on parallel processing architectures

O.-C. Granmo, Finn Verner Jensen

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearch

Abstract

Feature extraction in content-based indexing of media streams is often computational intensive. Typically, a parallel processing architecture is necessary for real-time performance when extracting features brute force. On the other hand, Bayesian network based systems for hypothesis driven feature extraction, which selectively extract relevant features one-by-one, have in some cases achieved real-time performance on single processing element architectures. In this paperwe propose a novel technique which combines the above two approaches. Features are selectively extracted in parallelizable sets, rather than one-by-one. Thereby, the advantages of parallel feature extraction can be combined with the advantages of hypothesis driven feature extraction. The technique is based on a sequential backward feature set search and a correlation based feature set evaluation function. In order to reduce the problem of higher-order feature-content/feature-feature correlation, causally complexly interacting features are identified through Bayesian network d-separation analysis and combined into joint features. When used on a moderately complex object-tracking case, the technique is able to select parallelizable feature sets real-time in a goal oriented fashion, even when some features are pairwise highly correlated and causally complexly interacting.
Original languageEnglish
Title of host publicationProceedings of The 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02)
Publication date2002
Publication statusPublished - 2002
EventThe 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02) - Las Vegas, United States
Duration: 19 May 2010 → …

Conference

ConferenceThe 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02)
CountryUnited States
CityLas Vegas
Period19/05/2010 → …

Fingerprint

Feature extraction
Bayesian networks
Processing
Function evaluation

Cite this

Granmo, O-C., & Jensen, F. V. (2002). Real-time hypothesis driven feature extraction on parallel processing architectures. In Proceedings of The 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02)
Granmo, O.-C. ; Jensen, Finn Verner. / Real-time hypothesis driven feature extraction on parallel processing architectures. Proceedings of The 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02). 2002.
@inproceedings{0777e1809c2d11db8ed6000ea68e967b,
title = "Real-time hypothesis driven feature extraction on parallel processing architectures",
abstract = "Feature extraction in content-based indexing of media streams is often computational intensive. Typically, a parallel processing architecture is necessary for real-time performance when extracting features brute force. On the other hand, Bayesian network based systems for hypothesis driven feature extraction, which selectively extract relevant features one-by-one, have in some cases achieved real-time performance on single processing element architectures. In this paperwe propose a novel technique which combines the above two approaches. Features are selectively extracted in parallelizable sets, rather than one-by-one. Thereby, the advantages of parallel feature extraction can be combined with the advantages of hypothesis driven feature extraction. The technique is based on a sequential backward feature set search and a correlation based feature set evaluation function. In order to reduce the problem of higher-order feature-content/feature-feature correlation, causally complexly interacting features are identified through Bayesian network d-separation analysis and combined into joint features. When used on a moderately complex object-tracking case, the technique is able to select parallelizable feature sets real-time in a goal oriented fashion, even when some features are pairwise highly correlated and causally complexly interacting.",
author = "O.-C. Granmo and Jensen, {Finn Verner}",
year = "2002",
language = "English",
booktitle = "Proceedings of The 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02)",

}

Granmo, O-C & Jensen, FV 2002, Real-time hypothesis driven feature extraction on parallel processing architectures. in Proceedings of The 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02)., Las Vegas, United States, 19/05/2010.

Real-time hypothesis driven feature extraction on parallel processing architectures. / Granmo, O.-C.; Jensen, Finn Verner.

Proceedings of The 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02). 2002.

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearch

TY - GEN

T1 - Real-time hypothesis driven feature extraction on parallel processing architectures

AU - Granmo, O.-C.

AU - Jensen, Finn Verner

PY - 2002

Y1 - 2002

N2 - Feature extraction in content-based indexing of media streams is often computational intensive. Typically, a parallel processing architecture is necessary for real-time performance when extracting features brute force. On the other hand, Bayesian network based systems for hypothesis driven feature extraction, which selectively extract relevant features one-by-one, have in some cases achieved real-time performance on single processing element architectures. In this paperwe propose a novel technique which combines the above two approaches. Features are selectively extracted in parallelizable sets, rather than one-by-one. Thereby, the advantages of parallel feature extraction can be combined with the advantages of hypothesis driven feature extraction. The technique is based on a sequential backward feature set search and a correlation based feature set evaluation function. In order to reduce the problem of higher-order feature-content/feature-feature correlation, causally complexly interacting features are identified through Bayesian network d-separation analysis and combined into joint features. When used on a moderately complex object-tracking case, the technique is able to select parallelizable feature sets real-time in a goal oriented fashion, even when some features are pairwise highly correlated and causally complexly interacting.

AB - Feature extraction in content-based indexing of media streams is often computational intensive. Typically, a parallel processing architecture is necessary for real-time performance when extracting features brute force. On the other hand, Bayesian network based systems for hypothesis driven feature extraction, which selectively extract relevant features one-by-one, have in some cases achieved real-time performance on single processing element architectures. In this paperwe propose a novel technique which combines the above two approaches. Features are selectively extracted in parallelizable sets, rather than one-by-one. Thereby, the advantages of parallel feature extraction can be combined with the advantages of hypothesis driven feature extraction. The technique is based on a sequential backward feature set search and a correlation based feature set evaluation function. In order to reduce the problem of higher-order feature-content/feature-feature correlation, causally complexly interacting features are identified through Bayesian network d-separation analysis and combined into joint features. When used on a moderately complex object-tracking case, the technique is able to select parallelizable feature sets real-time in a goal oriented fashion, even when some features are pairwise highly correlated and causally complexly interacting.

M3 - Article in proceeding

BT - Proceedings of The 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02)

ER -

Granmo O-C, Jensen FV. Real-time hypothesis driven feature extraction on parallel processing architectures. In Proceedings of The 2002 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'02). 2002