Resolution-to-noise trade-off in linear image restoration

Author

Zervakis, Michael E. ; Venetsanopoulos, Anastasios N.

Author_Institution

Dept. of Comput. Eng., Minnesota Univ., Duluth, MN, USA

Volume

38

Issue

10

fYear

1991

fDate

10/1/1991 12:00:00 AM

Firstpage

1206

Lastpage

1212

Abstract

The incorporation of both spatial and spectral adaptivity in a linear restoration algorithm is addressed. The formulation of a combined criterion is proposed, involving the minimum-mean-square-error (MMSE) and the least-mean-square-error (LMSE) criteria, in portions controlled by an indicator of the spatial signal activity. The incorporation of spectral adaptivity is achieved through the use of a decorrelating matrix in each individual criterion. The restoration algorithm derived, called the resolution-to-noise trade-off (RNT) algorithm, offers the flexibility of applying either linear MMSE (Wiener) filtering, inverse filtering, or no filtering at all, depending on the indicator of the spatial signal activity and the decorrelating matrices. The relationship of the RNT algorithm to other linear noniterative restoration approaches is discussed. it is indicated that the RNT filter forms a generalization of restoration filters that involve the point-spread function (psf) in a linear fashion

Keywords

correlation theory; filtering and prediction theory; least squares approximations; picture processing; spectral analysis; RNT algorithm; decorrelating matrix; inverse filtering; least-mean-square-error; linear image restoration; minimum-mean-square-error; point-spread function; resolution-to-noise trade-off; spatial adaptivity; spatial signal activity; spectral adaptivity; Decorrelation; Filtering; Humans; Image resolution; Image restoration; Noise level; Nonlinear filters; Signal restoration; Spatial resolution; Wiener filter;

fLanguage

English

Journal_Title

Circuits and Systems, IEEE Transactions on

Publisher

ieee

ISSN

0098-4094

Type

jour

DOI

10.1109/31.97540

Filename

97540