FLANN-based adaptive threshold selection for detection of impulsive noise in images.pdf

Int. J. Electron. Commun. (AEü) 61 (2007) 478–484 www.elsevier.de/aeue LETTER FLANN-based adaptive threshold selection for detection of impulsive noise in images Banshidhar Majhi?, Pankaj Kumar Sa Department of Computer Science and Engineering, National Institute of Technology Rourkela, Rourkela 769008, Orissa, India Received 13 February 2006 Abstract In this paper, a novel scheme has been suggested for removing random-valued impulsive noise from images. The proposed scheme utilizes a second-order differential impulse detection followed by a recursive median filter on the corrupted pixel locations. Adaptive threshold selection from noisy image characteristics has been emphasized in this paper. A functional link artificial neural network is used for this purpose. Comparative analysis on standard images at different noise conditions shows that the proposed scheme, in general, outperforms the existing schemes.  2006 Elsevier GmbH. All rights reserved. Keywords: Random-valued impulsive noise; Impulse detection; Median filter; Threshold selection; Sliding window; Adaptive threshold 1. Introduction Images often get contaminated with impulsive noise dur- ing acquisition and transmission. It is characterized by short, abrupt alterations of the gray levels in the image. Effective removal of noise while keeping the image features intact is one of the major problems in the field of computer vision and image processing [1]. Averaging filters have low-pass characteristics and when used to suppress impulsive noise they tend to blur the edges and other fine image details. However, nonlinear filtering techniques [2,3] are reported to produce satisfactory results. Median filter [4] is one of the most popular nonlinear techniques because of its sim- plicity in implementation and efficiency in noise removal. Weighted Median (WM) filters [5–7] are the variations of simple median filter, where some pixels are provided with more weight than other pixels in the test window. Cen- ter Weig