An Efficient Random Valued Impulse Noise Suppression Technique Using Artificial Neural Network and Non-Local Mean Filter

An Efficient Random Valued Impulse Noise Suppression Technique Using Artificial Neural Network and Non-Local Mean Filter

Bibekananda Jena (Anil Neerukonda Institute of Technology & Sciences, Visakhapatnam, India), Punyaban Patel (Malla Reddy Institute of Technology, Secunderabad, India) and G.R. Sinha (CMR Technical Campus, Secunderabad, India)
Copyright: © 2018 |Pages: 16
DOI: 10.4018/IJRSDA.2018040108

Abstract

A new technique for suppression of Random valued impulse noise from the contaminated digital image using Back Propagation Neural Network is proposed in this paper. The algorithms consist of two stages i.e. Detection of Impulse noise and Filtering of identified noisy pixels. To classify between noisy and non-noisy element present in the image a feed-forward neural network has been trained with well-known back propagation algorithm in the first stage. To make the detection method more accurate, Emphasis has been given on selection of proper input and generation of training patterns. The corrupted pixels are undergoing non-local mean filtering employed in the second stage. The effectiveness of the proposed technique is evaluated using well known standard digital images at different level of impulse noise. Experiments show that the method proposed here has excellent impulse noise suppression capability.
Article Preview

Introduction

Image denoising is one of the widely studied unsolved problems and plays a significant role in the research area of image processing and computer vision. Most of the time images are contaminated by impulse noise during the process of image acquisition or at the time of transmission due to malfunctioning image pixels in the camera sensors, channel transmission errors or faulty storage hardware. Therefore, a pre-processing stage is always required before processing an image for any application. Noise filtering is one of the important parts of this stage. The objectives of image denoising algorithms are to detect and suppress the unhealthy pixel elements in the test image without harming the fine details of the image. Impulse noise found in digital images is a spark that disturbs the information contain in the images. It distorts the pixels of a digital image by replacing the original value either by fixed value or any random value within the available dynamic range. So there are two categories of impulse noise as per the distribution of noise in an image: salt and pepper noise (SPN) and Random valued impulse noise (RVIN) (Dey, Ashour, Beagum et al., 2015) (Ikeda, Gupta, Dey et al., 2015). Impulse noises can be mathematically described by the following model (Patel, Jena, Majhi & Tripathy, 2012):

(1) where, specifies a certain location of a pixel in the test noisy image, indicates an original healthy image pixel element and represents a location where a pixel element is contaminated by impulse noise. In salt-and-pepper noise(SPN), the corrupted pixels either obtain the minimum value or the maximum value of the available dynamic range i.e. , and in random-valued impulse noise(RVIN), the corrupted pixel element can attain any possible value of the available dynamic range i.e. where, denote the minimum and the maximum pixel intensity within the available dynamic range of the image respectively. Therefore, the detection and suppression of presence of RVIN in an image is a challenging work (Gonzalez & Wood, 2002; Chanda & Majumder, 2002; Kang & Wang, 2009). Figure 1 shows the result of image corrupted by RVIN and SPN noise. Most of the existing algorithm performs very well for suppression of salt and pepper noise (SPN), whereas the performance towards random valued impulse noise (RVIN) is quite miserable (Jena, Patel &Majhi, 2014).

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 6: 4 Issues (2019): Forthcoming, Available for Pre-Order
Volume 5: 4 Issues (2018): 2 Released, 2 Forthcoming
Volume 4: 4 Issues (2017)
Volume 3: 4 Issues (2016)
Volume 2: 2 Issues (2015)
Volume 1: 2 Issues (2014)
View Complete Journal Contents Listing