Characterization of Transients.

Abstract

We introduced a signal representations to defect and analyze transients. A new algorithm called matching pursuit was developed to decompense any signal as a sum of waveforms that are chosen adaptively from a redundant dictionary of patterns, in order to best match the signal structures. This algorithm was applied to dictionary of dilated Gabor functions in order to characterize oscillatory transients of various sizes and frequencies. The asymptotic properties of this algorithm have been analyzed and we proved the existence of an attractor. This led to a general noise removal procedure which has been applied to audio signals. A fast matching pursuit algorithm was also designed and implemented in a software that is freely available on the internet. The matching pursuit algorithm has been extended in two dimensions for image processing. The image dictionary is composed of translated, dilated, and rotated wavelets. Applications to texture discrimination have been studied. To isolate patterns whose support may intersect, we have introduced a high resolution pursuit algorithm which was used to decompose high resolution radar signals.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Dec 31, 1995
Accession Number
ADA311108

Entities

People

  • Stephane Mallat

Organizations

  • New York University

Tags

Communities of Interest

  • Energy and Power Technologies
  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Algorithms
  • Applied Mathematics
  • Computational Complexity
  • Computations
  • Computer Science
  • Covariance
  • Detection
  • Equations
  • Frequency
  • High Resolution
  • Image Processing
  • Mathematics
  • Noise
  • Radar Signals
  • Stationary Processes
  • Two Dimensional
  • Waveforms

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Computer Vision.
  • Vision Science/Vision Psychology/Cognitive Neuroscience.