Compressive Imaging via Approximate Message Passing

Abstract

This project considers compressive imaging problems, where images are reconstructed from as few linear measurements as possible. Compressive imaging can be applied to a broad range of applications, including medical imaging, seismic imaging, and hyperspectral imaging. We propose novel compressive imaging algorithms that employ approximate message passing (AMP), which is an iterative signal estimation algorithm that performs component-wise denoising to noisy signals. In contrast, we apply non-separable denoisers to imaging with random matrices and hyperspectral imaging in CASSI systems. Numerical results demonstrate that our proposed algorithms significantly improve over current state-of-the-art compressive imaging algorithms in terms of both estimation error and run-time.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 04, 2015
Accession Number
AD1001362

Entities

People

  • Dror Baron

Organizations

  • North Carolina State University

Tags

Communities of Interest

  • Biomedical
  • Human Systems

DTIC Thesaurus Topics

  • Abstracts
  • Acquisition
  • Agreements
  • Algorithms
  • Coefficients
  • Compressed Sensing
  • Data Compression
  • Department Of Defense
  • Engineering
  • Hyperspectral Imagery
  • Mathematics
  • Observation
  • Signal Processing
  • Students
  • Technology Transfer
  • Three Dimensional
  • Two Dimensional

Readers

  • Approximation Theory.
  • Image Processing and Computer Vision.
  • Parallel and Distributed Computing.