1.1 Overview
A person receives maximum information about an object or a living being through images. An image is an illustration or general imprint of an object. It can also be defined as a two variable function o(i,j) where for each position (i,j) in the projection plane o(i,j) defines the light intensity at that point. The most commonly used types of images are binary image, gray image and color image. Binary images contain only black and white colors, also known as one-bit images. Images which have only brightness information and grayscale intensity are called gray images. They contain 8-bit data which implies 256 brightness levels. A 0 is used to represent black while 255 is used for white. Color images are those that contain three band monochrome information. These bands contain the brightness level information. Color image is comprised of picture elements called as pixels and the pixel is represented by a vector o(i,j) for a particular location, which has three intensity values o1(i,j), o2(i,j) and o3(i,j) each corresponding to red, green and blue colors, respectively (Plataniotis & Venetsanopoulos, 2000; Gonzalez & Woods, 2018; Petrou & Petrou, 2010). In the present day, visual information transferred in the form of digital images is becoming a primary medium of communication. The received image needs processing before it can be used in various applications like face recognition, surveillance, medical imaging, robot vision, underwater imaging, satellite imaging, remote sensing (Pal & Biswas, 2009; Dubey & Katarya, 2021; Ashok, 2021) etc. Frequently, the received image is of low quality due to problems such as noise, poor brightness, contrast, blur or artefacts. Image processing is a branch of engineering that investigates ways for restoring a damaged image to its original state. Image denoising (the reduction of noise from images) is primary pre-processing task for image analysis methods because noise is an unwanted and unavoidable component that is mixed with the original image in a variety of situations, such as during image acquisition, storage and transmission. Noise can highly dilute the image quality as it occurs due to multiple sources such as the transmission of image, dust on the camera lens, faulty photo sensors and faulty memory locations (Julliand et al., 2016). Generally, faulty photo sensors and faulty memory locations cannot be avoided as these occur due to the aging of electronic components. The possible types of noise that can affect images are: Gaussian noise, Shot noise, Impulse noise, Speckle noise, Thermal noise, etc. Gaussian noise originates from thermal vibration of atoms and discrete nature of radiation. Shot noise is a noise that occurs due to discrete nature of light. Impulse noise is the one type of noise which randomly modifies the pixel values and can be classified into fixed valued or SPN and RVIN. The pixel values get modified in case of SPN by only two values, either high or low value of the range whereas in case of RVIN the pixel values get modified independently as well as randomly. Speckle noise comes under the category of multiplicative noise which when introduces in any image then it is multiplied with the true pixel value of the noise free image. Thermal noise arises due to thermal energy of the chip. The effect of SPN on an image is shown in Figure 1a and noise reduction to get the denoised image is shown in Figure 1b.
Figure 1. Effect of SPN (a) Noisy image (b) Denoised image
Every day a massive number of images are captured and stored, but both these tasks are prone to noise. Because these images are regarded as a crucial source of information, a large number of them are communicated, stored and analyzed. Any loss of image information might have a negative impact on the overall performance of the system that contains the image processing step. So, day by day the demand for more conspicuous and accurate images is increasing. To fulfill this demand noise is required to be removed from the images. The most prevalent type of image noise is impulse noise, which may form any pattern, making it even more difficult to locate the source of the noise and forecast the original value of the noisy pixel. This is a basic problem in digital image processing, yet it continues to capture the attention of diverse academics since the requirement for improved image visual clarity is constantly in demand. Last few decades various methods have been developed for denoising of color images such as based upon Median Filter (Ko & Lee, 1991; Astola et al., 1990; Dong & Xu, 2007; Toh et al., 2008; Kang & Wang, 2009; Toh & Isa, 2009; Wang et al., 2010; Nair & Raju, 2012; Nair & Mol, 2013; Jin et al., 2008; Xu et al., 2014; Li et al., 2014; Jin et al., 2016; Hung & Chang, 2017; Roy et al., 2017; Zhu et al., 2018; Erkan et al., 2018; Erkan & Gokrem, 2018; Chen et al., 2019; Taha & Ibrahim, 2020; Jin et al., 2019; Noor et al., 2020; Erkan et al., 2020; Gupta et al., 2015; Smolka & Chydzinski, 2005; Celebi & Aslandogan, 2008; Zhao et al., 2012; Gellert & Brad, 2016; Erkan & Kilicman, 2016; Roig & Estruch, 2016; Hwang & Haddad, 1995; Sreenivasulu & Chaitanya, 2014; Sun et al., 2015; Lu & Chou, 2012; Yin et al., 1996; Arce, 1998; Arce & Paredes, 2000; Pattnaik et al., 2012; Palabaş & Gangal, 2012; Yu & Lee, 1993; Tsirikolias, 2016; Chen et al., 1999; Habib et al., 2015; Malinski & Smolka, 2019; Smolka & Malinski, 2018; Habib et al., 2015; Chen et al., 2020; Sa & Majhi, 2010; Singh et al., 2020; Geng et al., 2012; Wang et al., 2014; Jin et al., 2011), Fuzzy Logic (Wang et al., 2015; Habib et al., 2016; Roy et al., 2018; Xiao et al., 2016; Schulte et al., 2006; Schulte et al., 2007; Masood et al., 2014; Astola & Kuosmanen, 2020; Singh & Verma, 2021; Xiao et al., 2011; Jin et al., 2012), Principal Component Analysis (PCA) (Zhang et al., 2010; Dai et al., 2017), Anisotropic Diffusion (AD) (Xu et al., 2016; Jiranantanagorn, 2019), Optimization (Kumar et al., 2017; Khaw et al., 2019), Neural Networks (NN) (Li et al., 2020; Islam et al., 2018; Turkmen, 2016; Zhang et al., 2020), etc. and it is very difficult to choose one method for the desired application. This demands more technological advancements in image denoising methods to maintain image quality especially in color images. The problem imposed by impulse noise is more challenging for color images as there are three channels for noise reduction and the other most prominent distortion involved in color image deterioration is color artifacts. Due to these inevitable challenges, image denoising in color images is still a significant field and demanding for constant improvement. Hence it will be a challenging task to find a suitable method for color images.