Article Preview
Top1. Introduction
Noise will be introduced in a natural image during acquisition and transmission. The noise added during acquisition is generally related to the acquisition device faults and misalignments. Though, there is a significant amount of noise that is being added during the acquisition, the total noise is generally attributed to the channel for better interpretation. Compared to the noise during acquisition, the noise that is added during transmission is non-deterministic. Denoising is the activity of retrieving the best estimate of original image from the noisy image. This kind of retrieval is possible and become easy when the noise distribution is known to the denoising process. Additive White Gaussian Noise (AWGN) is one of the popular noise kinds that arises due to electron motion with thermal fluctuations in electronic circuits (Li, R., & Zhang, Y. J. 2003). Impulsive Noise (IN) is another important image noise type that arises duo to memory location errors in memory system, camera sensor pixels and transmission bit errors (Yan, M. 2013). Many works are available in the literature that treat either AWGN or Impulsive noise (Ko et al., 1991), (Youssef et al., 2015) (Kim et al., 2020), (Xu et al., 2020), (Jung et al. 2020) (Chen et al. 2001) (Sunkara et al., 2013) (Dong et al., 2007) (Zhong et al., 2021) (Xiong et al., 2011). A noise source can’t be of a single type. A mixed noise which a combination of large number of noise sources of same and non-same type is very common in practice. In addition to these schemes, schemes that handle mixed noise are also proposed in the literature (Rodríguez et al., 2012) (Dong, B et al., 2012) (Liu, J et al., 2012) (Zhuang et al., 2020) (Jiang, J et al., 2014) (Pitas, I. 1990).
As mentioned earlier, impulsive noise introduces pixel variations. Median filters and other non-linear filtering techniques are widely used to handle the impulsive noise (Sunkara, J. K et al., 2017). The median filtering does not tend to detect noisy pixels and applied the denoising to each pixel. Apart from this, the median filtering destroys the local structures of the image. This effect increases in proportion with the density of impulsive noise.
Modifications of median filtering like weighted median filtering Brownrigg, D. R. (1984), center weighted median filtering (Youssef et al., 2015) and multi-state median filtering (Hwang, H. et al., 1995) perform denoising by not considering whether a pixel is corrupted by impulsive noise. The better alternative way of performing denoising is to detect the pixels corrupted by noise and then process these pixels and leave the remaining pixels as it is. These schemes are termed as representative schemes in the literature. Representative median filtering schemes like adaptive median filtering (Sun, T. et al., 1994), switching median filtering (Sunkara, J. K et al.,) and directional weighted median filtering detects the noisy pixels and process them.