A Novel Approach for Edge Detection in Images Based on Cellular Learning Automata

Cellular Learning Automata (CLA) has been used in many fields of image processing such as noise elimination, smoothing, retrieval, fractionated and extraction of the content Characteristics of the images. The edge detection in images and methods if edge detection, have a great role in machine vision and cognizance systems. This method uses operands for analyzing images and digital image processing. Many studios here been conducted till now in edge detection algorithms of various conditions. In this study a new method for edge detection in images with the use of CLA is recommended. The proposed method of edge detection in images was tested with different sizes and the results were compared with Sobel edge detector classic method. The result show that the method based on CLA has a desirable performance in edge detection and compares the images with a more uniformity during a minimum period of time. DOI: 10.4018/ijcvip.2012100105 52 International Journal of Computer Vision and Image Processing, 2(4), 51-61, October-December 2012 Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. images with little background successful and complete edge detection is too hard (Maini & Aggarwal, 2007). The algorithms which are presented for edge detection in images have two main problems [3, 4]. One is that due to the presence of noise in images, the points which are not edges, are detected as edges and the other one is that the quality of the original image is low that causes the boundary of the objects to be undetectable. In this paper each image cell is considered as a different structure in automata and each learning automata has two state edges or non-edges. The local rules of cellular learning automata in repetitive continues methods are defined with edge and non-edge pixels and noises. In this paper, the new method of edge detection based on CLA is compared with other edge detection methods. The rest of the paper is organized as following: section 2 is a brief review of the related works and section 3 a brief description about CLA is presented. In section4, the process of edge detection based on CLA is presented. Section 5 includes examination and the results of simulation. Finally, section 6 of the paper involves conclusion. 2. LITERATURE REVIEW In an image, the distance between an object and background or between overlapped objects is called edge. Assuming that each image has uniform light intensity, the light intensity of the adjacent objects have different amounts and any variation in light intensity is considered linear filters results in ease of edge detection and decrease in calculations in order to get desired results. In the reference Pithadiya (2009) ̨ edge extraction concepts and edge detection processes in canny algorithm as well as morphological algorithm have been reviewed, the practical results of forenamed algorithms for clarity of edges in aerial photography of different areas of the city have been investigated. Ke et al. (2008) presented the cloud model is used for analyzing the edge information of the images in evolutionary modes of cell automata, Also it is used for finding the relationship between the adjacent pixels. Abin et al. in 2009 proposed a new method which was presented for segmenting the color images by using soft and hard segmenting processes. The process of soft segmentation of images and provides the threshold mode till it reaches the final segmentation and in hard segmentation, CLA analysis are done on input image and pixels of each part of the image. The algorithms presented to edge detection (Michael, Sudeep, Thomas, & Kevin, 1997; Abin, Fotouhi, & Kasaei, 2009; Abin, Fotouhi, & Kasaei, 2008). Are classic methods of edge detection such as Robert, Sobel which are calculates the gradient local maximum in the domain of location. If pixels are placed on the picture borders, their adjacency will be on graylevels. For detecting edges on crossover areas in Laplace conversion domain, the crossing points of second derivative of image function is considered as edges. In Abin, Fotouhi, and Kasaei (2009) and Abin, Fotouhi, and Kasaei (2008) canny method, edge detection and noise elimination from image, is defined by Gaussian function in which points with maximum gradient are delineated and the rest of the points are elimination from probability distribution. Chang et al. (2004) and Mirzai et al. (2011) investigated the edge detection using cellular automata for measuring the information or scaling of image and comparing the scale of previous matrix without people’s intervention, and they investigated the edge detection based on cloud model of cellular automata. In cloud model of cellular automata, the specific information of the edge and edge properties are discussed with using the information resulted from edge detection which is resulted from evolutionary cellular automata. The algorithm of edge detection on images using learning cellular automata is implemented through the use of original information of image and edge after receiving feedback from input data. In the reference Meybodi and Enayatifar (2009), internal image is an evolutionary model International Journal of Computer Vision and Image Processing, 2(4), 51-61, October-December 2012 53 Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. of original image, and external image detects the edge with the best quality and creates a background o the original image. Abin et al. (2008) proposed a fuzzy cellular automaton which is used in gray surfaces the images for noise elimination. In this method specific adjacency status is allocated for each pixel. The fuzzy logic recognizes the status of each pixel precisely, and in this recommended method, the edge detection is implemented by various instructions, and edge is detected in noised images. 3. LEARNING CELLULAR AUTOMATA 3.1. Cellular Automata Cellular automata (Koza, 1993; Thomas, 2012; Zhang, Zhong, & Zhao, 2007) was posed by Neuman (Zhang, Zhong, & Zhao, 2007; Neuman, 1966) in the late 1940s and after him it was proposed by mathematician Ulam (Ulam, 1972; Zhang, Zhong, & Zhao, 2007) as a model of producing calculations and simulation of systems in which multiple simple components cooperate for producing more complex patterns. Cellular automata are discrete dynamic systems that the operation is as local relations. In this model space is defined as a network in which each part is called a cell. And it is composed of a limited number of cells. For example two simple components and One-dimensional cellular automata are considered as a liner cell and each of the cells takes either zero or one. The system status in cellular automata is specified with using the total cell status the relationship between adjacencies and environmental rules (Zhang, Zhong, & Zhao, 2007; Cheng-hu, Zhan-li, & Yi-chun, 1999). The environment rules, includes either the simulation of definitive cellular automata rules or probable cellular automata, which is a set of local cellular updating. The updating of all of the cells is done according to the previous status of adjacent cell and not according to the present cell status. The locality means that each cell obtains its new status considering the status of adjacent cells. In cellular automata space and time are discrete. The problems of cellular automata are its inability in specifying the definite form of rules because of presence of noise in some systems and its indefiniteness, which is a hard and impossible work. 3.2. Learning Automata Learning automata (Nooraliei & Altun, 2009; Bwigy & Mybodi, 2010; Thathachar & Sastry, 2002) is machines which have a limited number of probabilities and for each of the learning automata a probability vector is allocated. This vector determines that each operation be implemented with what probability. Each selected operation is estimated by probability environment and the result is given to automata as a positive and negative signal. The next operation automaton chooses the best operation. The best operation is the one in which the probability of receiving reward from the environment is higher. In Figure 1 the function of learning automata in connected with environment in a feedback loop is shown. The environment is in the form of E = { } α β , , c , in which α α α α = ... { , , , } 1 2 r is the set of inputs and β β β β = ... { , , , } 1 2 r is the set of outputs and c c c cr = ... { } 1 2 , , , is the set of unsuccessful probabilities. When β of the environment is considered as a binary set, if the response of the environment equals zero, it means the reward mode, and if not, it is considered as the punishment mode (Thathachar & Sastry, 2002); Dong-Su, Wang-Heon, & InSo, 2004). The learning automaton is derided to two types according to the structure. The learning automata with fixed structure is shown as five mode of α β , , , , f g ∅ { } in which α = ... { , , , } α α α 1 2 r is the set of operations of automata { , , , } β = ... β β β 1 2 r , is the set of input data of automata,∅ = ∅ ∅ ... ∅ { , , , } 1 2 r is the set of internal status of automata of is the product function resulted from the new status of automata and is the output function which announces the automata status to the next out54 International Journal of Computer Vision and Image Processing, 2(4), 51-61, October-December 2012 Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. put. The learning automata is shown by fore m o d e o f { , , , } α β p T , i n w h i c h α = ... { , , , } α α α 1 2 r is the set of operations of automata β = ... { , , , } β β β 1 2 r , is the set of input data of automata , { , , , } = ... p p p r 1 2 i s t h e s e l e c t i o n p r o b a b i l i t y v e c t o r, p n T n n p n + ( ) = ( ) ( ) ( ) 1 [ , , ] α β is the learning algorithm reached by selection probability vector of the next operation. This type of automata is shown by linear learning algorithms (Meybodi & Enayatifar, 2009; Thathachar & Sastry, 2002; Dorigo & Gambardella, 1997). Learning automata can be used in solving same problems of optimization such as Ant Colony Optimization (ACO) (Dorigo & Gambardella, 1997) and Travelling salesman problem (TSP) (Alipor, 2012) and Network Routing Web Communities (Larranaga, Etxeberria, Lozano, & Pena, 1999).