Image restoration using L1-norm regularization and a gradient-based neural network with discontinuous activation functions

Author

Ferreira, Leonardo V. ; Kaszkurewicz, Eugenius ; Bhaya, Amit

Author_Institution

Dept. of Electr. Eng., Fed. Univ. of Rio de Janeiro, Rio de Janeiro

fYear

2008

fDate

1-8 June 2008

Firstpage

2512

Lastpage

2519

Abstract

The problem of restoring images degraded by linear position invariant distortions and noise is solved by means of a L₁-norm regularization, which is equivalent to determining a L₁-norm solution of an overdetermined system of linear equations, which results from a data-fitting term plus a regularization term that are both in L₁ norm. This system is solved by means of a gradient-based neural network with a discontinuous activation function, which is ensured to converge to a L₁-norm solution of the corresponding system of linear equations.

Keywords

gradient methods; image restoration; neural nets; L₁-norm regularization; data fitting; discontinuous activation function; gradient-based neural network; image restoration; linear position invariant distortion; noise; Degradation; Equations; Focusing; Helium; Image restoration; Image sensors; Inverse problems; Neural networks; Optical distortion; Predistortion;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks, 2008. IJCNN 2008. (IEEE World Congress on Computational Intelligence). IEEE International Joint Conference on

Conference_Location

Hong Kong

ISSN

1098-7576

Print_ISBN

978-1-4244-1820-6

Electronic_ISBN

1098-7576

Type

conf

DOI

10.1109/IJCNN.2008.4634149

Filename

4634149