Multilevel Thresholding of Gel Electrophoresis Images using Firefly Algorithm M. H. Mohd Noor #1 , A. R. Ahmad #2 , Z. Hussain #3 , K. A. Ahmad #4 , Ainihayati A. R. *5 # Faculty of Electrical Engineering, Universiti Teknologi MARA Pulau Pinang Malaysia 1 halim5381@ppinang.uitm.edu.my * School of Biological Sciences, Universiti Sains Malaysia Malaysia 4 ainirahim@yahoo.com Abstract— Gel electrophoresis (GE) is a process of DNA, RNA and protein molecules separation using electric field applied to a gel matrix. This paper describes the image processing techniques applied on GE image to segment the bands from their background. A few pre-processing steps are applied on the image prior to the segmentation technique for the purpose of removing noise in the image. Multilevel thresholding using Otsu method based on Firefly Algorithm is developed. The experimental results show that the Otsu-FA produced good separation of DNA bands and its background. Keywords— Gel electrophoresis image (GE), Otsu method, Firefly Algorithm (FA), multilevel thresholding, DNA bands image. I. INTRODUCTION Gel electrophoresis is a tool that employs electric field applied to a gel matrix for separation of molecules. The separation occurs when negatively charged molecules travel towards the positive electrode when electric field is applied. Naturally, heavier and bigger molecules travel relatively slower compared to lighter and smaller molecules. Thus, the combination of heavier and lighter molecules produces DNA bands which can be used for genomic analysis. Normally, the gel is stained with dye such as Ethidium Bromide (EtBr) in order to make the bands fluoresce under UV light. The fluorescent images can be captured as digital images. The images may contain several vertical lanes in which each lane represents one sample. Each lane contains a number of horizontal bands and the position of each band contains information that is valuable to biologist. Typically, GE images are corrupted with noises, has smeared appearance and non-uniform background. Image processing techniques are necessary to enhance their quality. Numerous methods have been proposed and studied for GE image segmentation. Maramis and Delopoulos [1] modeled the background component by using fourth degree polynomial of two variables function. The estimated background intensities are subtracted from the images which will produce GE images with zero background intensity level. An attempt to estimate the parameters of the polynomial function that will minimize the sum of squared errors might leads to computationally expensive and time-consuming. Computation of variance and standard deviation of each column in images have been proposed to isolate the DNA bands from their background [2], [3]. But the success of the approach to extract all DNA bands is dependent on the quality of the image. If the image is very noisy and contains ambiguous pixels, the extraction process might fail. Global thresholding has been proposed and employed [4], [5], [6]. Thresholding can be classified as bi-level thresholding and multilevel thresholding. Bi-level thresholding can easily segment the bands from the background if a valley is clearly visible in the image histogram. The multilevel thresholding, based on the determined thresholds, pixels having gray levels within a specified range are grouped into one class. However, to determine exact locations of distinct valleys in a multimodal histogram of an image that can segment the image efficiently is not that simple and requires additional processing. In this paper, multilevel thresholding with Otsu method based on FA is proposed to segment GE images. FA is employed to speed-up the computation and obtains the optimum threshold levels selection. II. OTSU METHOD Otsu method as proposed by Otsu (1979) has been developed for bi-level thresholding can be described as follows [7]. Let ( ) y x, f denotes a 2D gray-level image with L grey levels and these gray levels are in the range ( ) [ ] 1 2, 1, 0, - … L , . The number of pixels with gray level i is denoted by i n . Then the probability of gray level i in an image is as follows. N n = p i i (1) Suppose the image is dichotomized into two classes 1 C and 2 C (background and foreground) at a threshold level, t. 1 C denotes pixels with level ( ) [ ] 1 2, 1, 0, - … t , and 2 C denotes pixels with level ( ) [ ] 1 - … L , t, . The means of the two classes, 1 C and 2 C are given as follows. 2011 IEEE International Conference on Control System, Computing and Engineering 978-1-4577-1642-3/11/$26.00 ©2011 IEEE 18