AN UNSUPERVISED NEURAL NETWORK FOR IMPULSIVE NOISE REMOVAL IN DIGITAL IMAGES

A., FAYED K.; A., MAHMOUD T.

doi:10.21608/asat.2001.31159

AN UNSUPERVISED NEURAL NETWORK FOR IMPULSIVE NOISE REMOVAL IN DIGITAL IMAGES

Document Type : Original Article

Authors

FAYED K. A. ¹
MAHMOUD T. A. ²

¹ Ph. D., Egyptian Armed Forces.

² M. Sc., Egyptian Armed Forces.

10.21608/asat.2001.31159

Abstract

ABSTRACT
Pictures or images play an important role as a mass communication medium. When images are coded and transmitted over noisy communication channels, images are often corrupted by impulse noise. A method is proposed to eliminate impulsive noise with gaussian or uniform distribution in digital images. This noise removing method is based on two steps: impulse noise detection and filtered image reconstruction. Motivated by the success of neural computing in pattern classification, an unsupervised neural network has been employed in detecting the positions of the noisy pixels. When the noisy pixels are detected, a number of noise-exclusive filtering algorithms are invoked to eliminate the noise. These filters do not affect those pixels that are not corrupted. The filtering scheme presented can suppress impulse noise effectively as well as avoiding blurring or degrading the digital image quality. Experimental results and associated statistics demonstrate that the performance of the noise-exclusive filters is superior to many other well-known methods.

Keywords