Properties of different estimates of the regularizing parameter for the least squares image restoration problem

Nikolas Galatsanos*, Aggelos K. Katsaggelos

*Corresponding author for this work

Research output: Contribution to journalConference articlepeer-review


Image restoration is a necessary step for higher level machine or human image analysis. In many applications image data are blurred and corrupted by additive noise. Because of the presence of singularities in the blurring operator, regularization is an effective method for obtaining satisfactory solutions to image restoration problems. The application of regularization necessitates a choice of a regularizing parameter which trades fidelity to the data with smoothness of the restored image. For most problems of interest the choice of the regularizing parameter is not known a priori. Methods based on the properties of the residuals and on the generalized cross-validation have been proposed for estimating the regularizing parameter. In this paper alternative methods are proposed to compute the regularizing parameter. The resulting values of the regularizing parameter are compared with the values resulting from the above mentioned methods. Furthermore, it is shown that under certain conditions all the above mentioned methods result in the same value for the regularizing parameter. Experimental results are presented which verify the previous theoretical results.

Original languageEnglish (US)
Pages (from-to)590-600
Number of pages11
JournalProceedings of SPIE - The International Society for Optical Engineering
StatePublished - Jan 1 1991
EventApplications of Optical Engineering: Proceedings of OE/Midwest '90 - Rosemont, IL, USA
Duration: Sep 27 1990Sep 28 1990

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Condensed Matter Physics
  • Computer Science Applications
  • Applied Mathematics
  • Electrical and Electronic Engineering


Dive into the research topics of 'Properties of different estimates of the regularizing parameter for the least squares image restoration problem'. Together they form a unique fingerprint.

Cite this