Evolving Cellular Automata for Detecting Edges in Hyperspectral Images B. Priego, F. Bellas, D. Souto, F. López-Peña, R. J. Duro Integrated Group for Engineering Research University of A Coruña Ferrol, Spain {blanca.priego, fran, dsouto, flop, richard} @ udc.es Abstract— This paper deals with the problem of segmenting or, more properly, finding edges in multidimensional images, in particular, hyperspectral images. The approach followed is based on the use of cellular automata (CA) and their emergent behavior in order to achieve this objective. Using cellular automata for finding edges in hyperspectral images is not new, but most current approaches to this problem involve hand designing the rules for the automata. On the other hand, many authors just use extensions of one-dimensional edge detection methods to multidimensional images, thus averaging out the spectral information present. Here, we consider the application of evolutionary methods to produce the CA rule sets that obtain the best possible edge detection properties under different circumstances and using spectral based approaches. The procedure has been tested over synthetic and real hyperspectral images and the results obtained have been compared to those produced using the hyper-Sobel and Hyper-Prewitt operators, which are standard edge detection methods for gray-level images that have been extended by some authors to the multidimensional domain. Keywords- Hyperspectral image processing; Cellular Automata; Evolutionary Algorithms. I. INTRODUCTION Hyperspectral imaging is becoming a very important source of remote sensing and industrial processing data with a very high level of spectral detail. Hyperspectral images differ from regular images in the fact that instead using three values to represent the color of each pixel (corresponding to the levels of red, green and blue), they use hundreds of values that correspond to the intensity of different bands of the visible and near infrared spectrum. This allows for a very accurate definition of color and, thus for a very high discrimination power. Typically, a hyperspectral image may consist of two spatial dimensions of anywhere from 250 pixels to a couple of thousand pixels and a spectral dimension of up to a thousand bands per pixel. Originally hyperspectrometers were mainly used in remote sensing applications and the images were obtained from high flying planes [1]. Analysis methods were developed to provide the ratio of endmembers present in every pixel so as to improve the spatial discrimination of these systems when analyzing different types of covers. In fact, the emphasis was placed on the spectral processing and not so much on the spatial or morphological features. Currently this is not the case, especially in ground based applications [2][3] where the images are taken close enough to the subject to obtain a relatively detailed view. Consequently, it is becoming more important to consider the geometric layout. In other words it is necessary to perform combined spatial-spectral processing in order to identify morphological features and this means segmenting objects from the background. A first step in this direction, and the problem we are addressing here, is being able to reliably detect edges within the images. The fine level of spectral and, currently, of spatial detail of hyperspectral images leads to very large data sets for each image. Processing that much data, especially if one wants to do this in real time, requires specialized techniques and a very high level of parallelism. The case of edge detection is no different and, as in other applications, the use of extremely distributed techniques, such as Cellular Automata, become very interesting and can provide the level of locality in computations that allows for the use of the current state of the art FPGA [4] or GPU [5] architectures. Cellular automata (CA) are a biologically inspired decentralized computing paradigm first proposed by Von Neumann and Ulam [6]. They can be described as a spatially extended decentralized system where a set of cells are distributed in space and communicate only with their neighbors. Each cell is basically an automaton (usually a finite state automaton) and the state of a cell each instant of time depends on the state of its neighbors and on its own previous state through a set of transition rules. It is by adequately choosing these rules and by iterating the state transition process in time that the full computational capabilities of the whole system are achieved. In fact, Von Neumann himself demonstrated that CAs can be made universal computing machines. As in many other realms, the challenge, especially when complex functions need to be achieved, is to determine the rules for the automata so that these functions are performed. In other words, one knows how the whole CA system has to behave and needs to find out what rules implemented in each cell will lead the system, as a whole, to behave that way and thus produce the desired result. This is what is called the inverse problem and it is a very difficult problem to solve. The authors within the CA community have resorted to different U.S. Government work not protected by U.S. copyright WCCI 2012 IEEE World Congress on Computational Intelligence June, 10-15, 2012 - Brisbane, Australia FUZZ IEEE