Segmentation with Learning Automata

Several image processing applications aim to detect and mark remarkable features which in turn might be used to perform high-level tasks. In particular, image segmentation seeks to group pixels within meaningful regions. Commonly, gray levels belonging to the object are substantially different from the gray levels featuring the background. Thresholding is thus a simple but effective tool to isolate objects of interest from the background. Its applications include several classics such as document image analysis, whose goal is to extract printed characters (Abak et al., 1997; Kamel & Zhao, 1993) logos, graphical content, or musical scores; also it is used for map processing which aims to locate lines, legends, and characters (Trier & Jain, 1995). It is also used for scene processing, aiming for object detection and marking (Bhanu, 1986); Similarly, it has been employed to quality inspection for materials (Sezgin & Sankur, 2001; Sezgin & Tasaltin, 2000), discarding defective parts. Thresholding selection techniques can be classified into two categories: bi-level and multilevel. In bi-level thresholding, one limit value is chosen to segment an image into two classes: one represents the object and the other represents the background. When an image is composed of several distinct objects, multiple threshold values have to be selected for proper segmentation. This is called multilevel thresholding. A variety of thresholding approaches have been proposed for image segmentation, including conventional methods (Guo & Pandit, 1998; Pal & Pal, 1993; Shaoo et al., 1988; Snyder et al., 1990) and intelligent techniques such as in (Chen & Wang, 2005; Chih-Chih, 2006). Extending the algorithm to a multilevel approach may arise some inconveniences: (i) they may have no systematic and analytic solution when the number of classes to be detected increases and (ii) the number of classes is either difficult to be predicted or must be pre-defined. However, this parameter is unknown for many real applications. In order to solve these problems, an alternative approach using an optimization algorithm based on learning automata for multilevel thresholding is proposed in this paper. In the traditional multilevel optimal thresholding, the intensity distributions belonging to the object or to the background pixels are assumed to follow some Gaussian probability function; therefore a combination of probability density functions is usually adopted to model these functions. The parameters in the combination function are unknown and the parameter estimation is typically assumed to be a nonlinear optimization problem (Gonzalez & Woods, 1990). The unknown parameters that give the best fit to the processed histogram are determined by using a LA algorithm (Thathachar & Sastry, 2002). The main motivation behind the use of LA as an optimization algorithm for parameter adaptation is to use its capabilities of global optimization when dealing to multimodal