Segmentation is one of the most important tasks in image processing. It classifies the pixels into two or more groups depending on their intensity levels and a threshold value. The classical methods exhaustively search the best thresholds for a spec image. This process requires a high computational effort, to avoid this situation has been incremented the use of evolutionary algorithms. The Electro-magnetism-Like algorithm (EMO) is an evolutionary method which mimics the attraction-repulsion mechanism among charges to evolve the members of a population. Different to other algorithms, EMO exhibits interesting search capabilities whereas maintains a low computational overhead. This chapter introduces a multilevel thresholding (MT) algorithm based on the EMO and the Otsu's method as objective function. The combination of those techniques generates a multilevel segmentation algorithm which can effectively identify the threshold values of a digital image reducing the number of iterations.
TopIntroduction
The success of an image processing system depends directly on the quality of the input images. Considering that the image acquisition process there exist different situations that a affects features of the scene, there is necessary to apply methods that permits the analysis of the elements contained in the image. Almost all methods of image processing require a first step called segmentation. This task permits the classification of pixels in the image depending on its gray (or RGB in each component) level intensity. Several techniques had been studied (Akay, 2012; Ghamisi, Couceiro, Benediktsson, & Ferreira, 2012; Hammouche, Diaf, & Siarry, 2010; Kapur, Sahoo, & Wong, 1985; Kittler & Illingworth, 1986; Liao, Chen, & Chung, 2001; Otsu, 1979; Sezgin & Sankur, 2004, Ali et al., 2015, Ibrahim et al., 2015). Thresholding is the easiest method for segmentation, it works taking a threshold (th) value and the pixels which intensity value is higher than are labelled as the first class and the rest of the pixels correspond to the second class. When the image is segmented in two classes it is called bi-level thresholding (BT) and it is necessary only one value. On the other hand, when pixels are separated in more than two classes it is called multilevel thresholding (MT) and there are required more than one values (Akay, 2012; Hammouche et al., 2010; J. N. Kapur, P. K. Sahoo, A. K. C. Wong, 1985; Kittler & Illingworth, 1986; Liao et al., 2001; Otsu, 1979; Sathya & Kayalvizhi, 2011; Sezgin & Sankur, 2004). Threshold based methods are divided in parametric and nonparametric (Akay, 2012; Hammouche, Diaf, & Siarry, 2008; Liao et al., 2001). The nonparametric employs a given criteria between-class variance, entropy and error rate (J. N. Kapur, P. K. Sahoo, A. K. C. Wong, 1985; Kittler & Illingworth, 1986; Otsu, 1979) that must be optimized to determine the optimal threshold values. These approaches result an attractive option due their robustness and accuracy (Sezgin & Sankur, 2004).
One of the classical method used for bi-level thresholding is the between classes variance and was proposed by Otsu (Otsu, 1979). The efficiency and accuracy have been already proved for two segmentation classes (Sathya & Kayalvizhi, 2011). Although Otsu’s can be expanded for multilevel thresholding, its computational complexity increases exponentially with each new threshold (Sathya & Kayalvizhi, 2011). This chapter introduces a multilevel threshold method based on the Electromagnetism-like Algorithm (EMO). EMO is a global optimization algorithm that mimics the electromagnetism law of physics. It is a population based and physical inspired method which has an attraction-repulsion mechanism to evolve the members of the population guided by their objective function values (Birbil & Fang, 2003).