Article Preview
TopIntroduction
Image compression is a technique of showing the images in procedural manner, that which reduce the number of bits essential to represent an image and in order to advance the capacity of the storage device. There are several techniques which are proposed by various researchers, but the most used image compression technique is Joint Photographic Expert Group (JPEG). Discrete Cosine Transformed (DCT) was firstly introduced later JPEG-2000 is introduced (Skodras et al., 2001).
There are many methods can be utilized for image compression, some of these methods depend on mathematical transforms such as; discrete cosine transform (Haweel et al., 2016), Discrete Wavelet Transforms (DWT) (Bruylants et al., 2015), Integer Wavelet Transforms (IWT) (Zhang and Tong, 2017), Karhunen Loeve Transforms (KLT) (Zhang & Tong, 2017), Hartley Transform (Sunder et al.,), Watershed Transform (Hsu, 2012), Walsh Hadamard Transform (WHT) (Venugopal et al., 2016), Tchebichef Transform Kiruba & Sumathy, and Singular Value Decomposition (SVD) (Kumar & Vaish, 2017).
Various clustering algorithms have been proposed for image compression and are categorized into two categories: hierarchical algorithm and partition algorithm (Jain et al., 2000). In Partition cluster, clusters are updated as per the smaller difference between the centroid and input data which is to be cluster. The entire input data is portioned into some clusters based on nearest distance between centroid and data. Whereas in hierarchical cluster, hierarchy (most important) is given to some clusters and are of two types: Agglomerative and Divisive. One can find detailed description of both in (Han et al., 2017).
Image Compression can also be done by non-transformed methods like Vector Quantization (VQ) and Thresholding. In image processing, selecting a gray level threshold from the images like gray level image/color and extracting its background image is a challenging task. Hence many techniques for distinguished gray level threshold are proposed by the researchers. Thresholding is generally applied because of its progressive robustness, accuracy and less time convergence. There are two ways of approaching thresholds, firstly parametric and the other is non- parametric. In non-parametric approach, depending on class variance thresholding it is performed as in Otsu technique or depending on the criterion of entropies like Shannon, Fuzzy and Kapur’s (De Luca & Termini, 1972). If in case, the image is divided into two classes which are object and background, then the threshold is called multi-level threshold or bi-level threshold. Thresholding techniques up hold various real time applications.
A detailed research on image thresholding was performed by Sezgin and Sankur in 2004 and classified it into six categories; those are Clustering-based, Entropy-based, Histogram shape-based, Object attribute-based, spatial and local methods respectively (Sezgin & Sankur, 2004). Considering the histogram’s of the gray level images, Kaur classified the images based on calculating threshold (Kaur et al., 2007). The image is portioned into desired classes based on optimizing the class variance and once example of such kind is Otsu’s method (Otsu, 1979). Under bi-level thresholding these two methods found effective in case of more than one threshold whereas in multi-level thresholding the complexity is high. The drawback of Bayesian error and Birge-Massart thresholding is that the computational or CPU time is exponentially rising with the problem and to overcome these problems (Rather & Bala, 2020), evolutionary and swarm-based calculation techniques are alternatives (Rather & Bala, 2020). Repositories of the codes are found in citations (Sajad Ahmad Rather, 2020), (Sajad Ahmad Rather, 2020) and (Seyedali Mirjalili, 2020)