Content area
Full Text
Abstract: Interpolation-Based Impulse Noise Removal (IBINR), a fast and simple algorithm is proposed to remove fixed valued impulse noise in this study. The proposed method removes all noisy pixels from the image and determines their values using non-linear interpolation. Compared with the state-of-the-art noise removal algorithms, IBINR has the highest or comparable performance in terms of photographic image denoising power and resource efficiency: it runs in shorter amount of time and does not have any significant additional memory requirements due to the fact that costly sorting operations are avoided.
(ProQuest: ... denotes formulae omitted.)
1 Introduction
Impulse noise is caused by many reasons, such as noisy channels or faulty hardware [1]. Impulse noise can either be fixed-valued (salt-and-pepper noise) or random-valued. General reason for random-valued impulse noise is the leaky pixels in image sensors [2]. Salt-and-pepper noise is a more common impulse noise type, because it can occur as a result of interference in addition to faulty hardware. In digital images, salt-and-pepper noise shows itself as either completely black or completely white pixels within the image [3]. For image-processing applications, these noisy pixels cause edge-like features; therefore, they should be removed before further processing [4]. Moreover, human eye is also sensitive to high-frequency distortions in the image. Many methods that aim to remove salt-and-pepper noise involve median filtering [5]. However, the most recent impulsive noise removal approaches use rather complex algorithms spanning multiple steps [6-9].
Our proposed interpolation-based impulse noise removal (IBINR) algorithm follows a different approach and utilises a simple method for noise removal. It incorporates non-linear interpolation to determine the actual values of noisy pixels, where a pixel is considered noisy if its intensity is the highest or the lowest possible value. While interpolating, noisy pixels are considered as missing; therefore, the proposed method uses the values that are known to be noise-free.
Computational requirements for salt-and-pepper noise removal are often overlooked and considered as the second priority goal [6] or are not even considered [8, 9]. However, salt-and-pepper noise removal might be required in hardware or in real-time situations. For instance, in outdoor surveillance systems, salt-and-pepper noise can be encountered [9]. An application that accepts image input from such devices may need to operate in real-time. In these situations, computational...