Contextually adaptive signal representation using conditional Principal Component Analysis

Rosa M Figueras I Ventura, Umesh Rajashekar, Zhou Wang, Eero P. Simoncelli

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The conventional method of generating a basis that is optimally adapted (in MSE) for representation of an ensemble of signals is Principal Component Analysis (PCA). A more ambitious modern goal is the construction of bases that are adapted to individual signal instances. Here we develop a new framework for instance-adaptive signal representation by exploiting the fact that many real-world signals exhibit local self-similarity. Specifically, we decompose the signal into multiscale subbands, and then represent local blocks of each subband using basis functions that are linearly derived from the surrounding context. The linear mappings that generate these basis functions are learned sequentially, with each one optimized to account for as much variance as possible in the local blocks. We apply this methodology to learning a coarse-to-fine representation of images within a multi-scale basis, demonstrating that the adaptive basis can account for significantly more variance than a PCA basis of the same dimensionality.

Original languageEnglish (US)
Title of host publication2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP
Pages877-880
Number of pages4
DOIs
StatePublished - 2008
Event2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP - Las Vegas, NV, United States
Duration: Mar 31 2008Apr 4 2008

Other

Other2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP
CountryUnited States
CityLas Vegas, NV
Period3/31/084/4/08

Fingerprint

principal components analysis
Principal component analysis
learning
methodology

Keywords

  • Adaptive basis
  • Conditional PCA
  • Image modeling
  • Image representation
  • Self-similarities

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Signal Processing
  • Acoustics and Ultrasonics

Cite this

I Ventura, R. M. F., Rajashekar, U., Wang, Z., & Simoncelli, E. P. (2008). Contextually adaptive signal representation using conditional Principal Component Analysis. In 2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP (pp. 877-880). [4517750] https://doi.org/10.1109/ICASSP.2008.4517750

Contextually adaptive signal representation using conditional Principal Component Analysis. / I Ventura, Rosa M Figueras; Rajashekar, Umesh; Wang, Zhou; Simoncelli, Eero P.

2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP. 2008. p. 877-880 4517750.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

I Ventura, RMF, Rajashekar, U, Wang, Z & Simoncelli, EP 2008, Contextually adaptive signal representation using conditional Principal Component Analysis. in 2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP., 4517750, pp. 877-880, 2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP, Las Vegas, NV, United States, 3/31/08. https://doi.org/10.1109/ICASSP.2008.4517750
I Ventura RMF, Rajashekar U, Wang Z, Simoncelli EP. Contextually adaptive signal representation using conditional Principal Component Analysis. In 2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP. 2008. p. 877-880. 4517750 https://doi.org/10.1109/ICASSP.2008.4517750
I Ventura, Rosa M Figueras ; Rajashekar, Umesh ; Wang, Zhou ; Simoncelli, Eero P. / Contextually adaptive signal representation using conditional Principal Component Analysis. 2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP. 2008. pp. 877-880
@inproceedings{75a6e740e67f4efeb49dd197fd5287eb,
title = "Contextually adaptive signal representation using conditional Principal Component Analysis",
abstract = "The conventional method of generating a basis that is optimally adapted (in MSE) for representation of an ensemble of signals is Principal Component Analysis (PCA). A more ambitious modern goal is the construction of bases that are adapted to individual signal instances. Here we develop a new framework for instance-adaptive signal representation by exploiting the fact that many real-world signals exhibit local self-similarity. Specifically, we decompose the signal into multiscale subbands, and then represent local blocks of each subband using basis functions that are linearly derived from the surrounding context. The linear mappings that generate these basis functions are learned sequentially, with each one optimized to account for as much variance as possible in the local blocks. We apply this methodology to learning a coarse-to-fine representation of images within a multi-scale basis, demonstrating that the adaptive basis can account for significantly more variance than a PCA basis of the same dimensionality.",
keywords = "Adaptive basis, Conditional PCA, Image modeling, Image representation, Self-similarities",
author = "{I Ventura}, {Rosa M Figueras} and Umesh Rajashekar and Zhou Wang and Simoncelli, {Eero P.}",
year = "2008",
doi = "10.1109/ICASSP.2008.4517750",
language = "English (US)",
isbn = "1424414849",
pages = "877--880",
booktitle = "2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP",

}

TY - GEN

T1 - Contextually adaptive signal representation using conditional Principal Component Analysis

AU - I Ventura, Rosa M Figueras

AU - Rajashekar, Umesh

AU - Wang, Zhou

AU - Simoncelli, Eero P.

PY - 2008

Y1 - 2008

N2 - The conventional method of generating a basis that is optimally adapted (in MSE) for representation of an ensemble of signals is Principal Component Analysis (PCA). A more ambitious modern goal is the construction of bases that are adapted to individual signal instances. Here we develop a new framework for instance-adaptive signal representation by exploiting the fact that many real-world signals exhibit local self-similarity. Specifically, we decompose the signal into multiscale subbands, and then represent local blocks of each subband using basis functions that are linearly derived from the surrounding context. The linear mappings that generate these basis functions are learned sequentially, with each one optimized to account for as much variance as possible in the local blocks. We apply this methodology to learning a coarse-to-fine representation of images within a multi-scale basis, demonstrating that the adaptive basis can account for significantly more variance than a PCA basis of the same dimensionality.

AB - The conventional method of generating a basis that is optimally adapted (in MSE) for representation of an ensemble of signals is Principal Component Analysis (PCA). A more ambitious modern goal is the construction of bases that are adapted to individual signal instances. Here we develop a new framework for instance-adaptive signal representation by exploiting the fact that many real-world signals exhibit local self-similarity. Specifically, we decompose the signal into multiscale subbands, and then represent local blocks of each subband using basis functions that are linearly derived from the surrounding context. The linear mappings that generate these basis functions are learned sequentially, with each one optimized to account for as much variance as possible in the local blocks. We apply this methodology to learning a coarse-to-fine representation of images within a multi-scale basis, demonstrating that the adaptive basis can account for significantly more variance than a PCA basis of the same dimensionality.

KW - Adaptive basis

KW - Conditional PCA

KW - Image modeling

KW - Image representation

KW - Self-similarities

UR - http://www.scopus.com/inward/record.url?scp=51449090515&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=51449090515&partnerID=8YFLogxK

U2 - 10.1109/ICASSP.2008.4517750

DO - 10.1109/ICASSP.2008.4517750

M3 - Conference contribution

AN - SCOPUS:51449090515

SN - 1424414849

SN - 9781424414840

SP - 877

EP - 880

BT - 2008 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP

ER -