Sparse representations for image decompositions

Davi Geiger, Tyng Luh Liu, Michael J. Donahue

Research output: Contribution to journalArticle

Abstract

We are given an image I and a library of templates L, such that L is an overcomplete basis for I. The templates can represent objects, faces, features, analytical functions, or be single pixel templates (canonical templates). There are infinitely many ways to decompose I as a linear combination of the library templates. Each decomposition defines a representation for the image I, given L. What is an optimal representation for I given L and how to select it? We are motivated to select a sparse/compact representation for I, and to account for occlusions and noise in the image. We present a concave cost function criterion on the linear decomposition coefficients that satisfies our requirements. More specifically, we study a 'weighted Lp norm' with 0 < p < 1. We prove a result that allows us to generate all local minima for the Lp norm, and the global minimum is obtained by searching through the local ones. Due to the computational complexity, i.e., the large number of local minima, we also study a greedy and iterative 'weighted Lp Matching Pursuit' strategy.

Original languageEnglish (US)
Pages (from-to)139-156
Number of pages18
JournalInternational Journal of Computer Vision
Volume33
Issue number2
DOIs
StatePublished - Sep 1999

Fingerprint

Decomposition
Cost functions
Computational complexity
Pixels

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Vision and Pattern Recognition
  • Control and Systems Engineering

Cite this

Sparse representations for image decompositions. / Geiger, Davi; Liu, Tyng Luh; Donahue, Michael J.

In: International Journal of Computer Vision, Vol. 33, No. 2, 09.1999, p. 139-156.

Research output: Contribution to journalArticle

Geiger, Davi ; Liu, Tyng Luh ; Donahue, Michael J. / Sparse representations for image decompositions. In: International Journal of Computer Vision. 1999 ; Vol. 33, No. 2. pp. 139-156.
@article{9a1ca00bf25a4940a5a7cfeba180612b,
title = "Sparse representations for image decompositions",
abstract = "We are given an image I and a library of templates L, such that L is an overcomplete basis for I. The templates can represent objects, faces, features, analytical functions, or be single pixel templates (canonical templates). There are infinitely many ways to decompose I as a linear combination of the library templates. Each decomposition defines a representation for the image I, given L. What is an optimal representation for I given L and how to select it? We are motivated to select a sparse/compact representation for I, and to account for occlusions and noise in the image. We present a concave cost function criterion on the linear decomposition coefficients that satisfies our requirements. More specifically, we study a 'weighted Lp norm' with 0 < p < 1. We prove a result that allows us to generate all local minima for the Lp norm, and the global minimum is obtained by searching through the local ones. Due to the computational complexity, i.e., the large number of local minima, we also study a greedy and iterative 'weighted Lp Matching Pursuit' strategy.",
author = "Davi Geiger and Liu, {Tyng Luh} and Donahue, {Michael J.}",
year = "1999",
month = "9",
doi = "10.1023/A:1008146126392",
language = "English (US)",
volume = "33",
pages = "139--156",
journal = "International Journal of Computer Vision",
issn = "0920-5691",
publisher = "Springer Netherlands",
number = "2",

}

TY - JOUR

T1 - Sparse representations for image decompositions

AU - Geiger, Davi

AU - Liu, Tyng Luh

AU - Donahue, Michael J.

PY - 1999/9

Y1 - 1999/9

N2 - We are given an image I and a library of templates L, such that L is an overcomplete basis for I. The templates can represent objects, faces, features, analytical functions, or be single pixel templates (canonical templates). There are infinitely many ways to decompose I as a linear combination of the library templates. Each decomposition defines a representation for the image I, given L. What is an optimal representation for I given L and how to select it? We are motivated to select a sparse/compact representation for I, and to account for occlusions and noise in the image. We present a concave cost function criterion on the linear decomposition coefficients that satisfies our requirements. More specifically, we study a 'weighted Lp norm' with 0 < p < 1. We prove a result that allows us to generate all local minima for the Lp norm, and the global minimum is obtained by searching through the local ones. Due to the computational complexity, i.e., the large number of local minima, we also study a greedy and iterative 'weighted Lp Matching Pursuit' strategy.

AB - We are given an image I and a library of templates L, such that L is an overcomplete basis for I. The templates can represent objects, faces, features, analytical functions, or be single pixel templates (canonical templates). There are infinitely many ways to decompose I as a linear combination of the library templates. Each decomposition defines a representation for the image I, given L. What is an optimal representation for I given L and how to select it? We are motivated to select a sparse/compact representation for I, and to account for occlusions and noise in the image. We present a concave cost function criterion on the linear decomposition coefficients that satisfies our requirements. More specifically, we study a 'weighted Lp norm' with 0 < p < 1. We prove a result that allows us to generate all local minima for the Lp norm, and the global minimum is obtained by searching through the local ones. Due to the computational complexity, i.e., the large number of local minima, we also study a greedy and iterative 'weighted Lp Matching Pursuit' strategy.

UR - http://www.scopus.com/inward/record.url?scp=0033188879&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0033188879&partnerID=8YFLogxK

U2 - 10.1023/A:1008146126392

DO - 10.1023/A:1008146126392

M3 - Article

VL - 33

SP - 139

EP - 156

JO - International Journal of Computer Vision

JF - International Journal of Computer Vision

SN - 0920-5691

IS - 2

ER -