4. Fast Bottom-Up Computational Models in the Spectral Domain

  1. Liming Zhang1 and
  2. Weisi Lin2

Published Online: 20 MAR 2013

DOI: 10.1002/9780470828144.ch4

Selective Visual Attention: Computational Models and Applications

Selective Visual Attention: Computational Models and Applications

How to Cite

Zhang, L. and Lin, W. (2013) Fast Bottom-Up Computational Models in the Spectral Domain, in Selective Visual Attention: Computational Models and Applications, John Wiley & Sons (Asia) Pte Ltd, Singapore. doi: 10.1002/9780470828144.ch4

Author Information

  1. 1

    Fudan University, P. R. China

  2. 2

    Nanyang Technological University, Singapore

Publication History

  1. Published Online: 20 MAR 2013
  2. Published Print: 27 MAR 2013

ISBN Information

Print ISBN: 9780470828120

Online ISBN: 9780470828144

SEARCH

Keywords:

  • spectral residual (SR);
  • phase spectrum of Fourier transform (PFT);
  • phase of quaternion Fourier transform (PQFT);
  • pulsed discrete cosine transform (PCT);
  • frequency domain division normalization (FDN);
  • patch FDN (PFDN);
  • pulsed PCA (principal components analysis);
  • amplitude spectrum of quaternion Fourier transform (AQFT);
  • Joint Photographic Experts Group (JPEG);
  • JPEG bit-stream;
  • human visual sensitivity;
  • frequency domain

Summary

This chapter continues the introduction to bottom-up visual attention models. Following the description of models in the spatial (pixel) domain in the previous chapter, the focus is now put on models in the spectral domain. Since frequency domain models can detect the salient object quicker to enable them to meet real-time requirements in engineering, they are the choice for many real-world applications

In this chapter, first the properties of the frequency spectrum for image analysis are given in Section 4.1, and then the major bottom-up computational models based on phase spectrum in frequency domain are presented in Sections 4.2–4.6: the SR, PFT, PQFT, PCT and FDN models, respectively. In Section 4.6, FDN and PFDN models have biological plausibility because they simulate each step from the (spatial domain) BS model, but in the frequency domain. In Section 4.7, the AQFT model based on amplitude spectrum of image patches is introduced and Section 4.8 gives a computational model from the JPEG bit-stream. Finally, the advantages and limitations of frequency computational models are discussed in Section 4.9.