Entropy-based Multilevel 2D Histogram Image Segmentation using DEWO Optimization Algorithm Garima Vig CS&E Department Amity University, U. P. NOIDA, India garimavig311@gmail.com MD. Shahbaz CS&E Department Amity University, U.P. Noida, India shahbaz.md5977@gmail.com Sapna Varshney Deptt. of Computer Science University of Delhi Delhi, India sapnavarsh@gmail.com Sumit Kumar CS&E Department Amity University, U.P. Noida, India sumitkumarbsr19@gmail.com Abstract—Thresholding is widely used image segmentation technique in many real-life applications like document image processing, quality inspection to detect defective parts of machines, medical imaging etc. Multilevel image segmentation is a simple approach for colored image segmentation with less computational complexity, but multilevel image segmentation is not able to properly exploit the spatial correlation of image's pixels. This study proposes a hybrid of Differential Evolution and Whale Optimization (DEWO) for entropy based multilevel image segmentation using non-local means 2Dhistogram and to perform colored image segmentation. The proposed approach is compared with some prominent meta heuristic algorithms in recent past using Tsallis entropy, Renyi entropy, and Kapur entropy functions to validate its efficiency for different entropy functions. Results obtained from the proposed approach for image segmentation is better than all the other meta-heuristic algorithms in every entropy-based segmentation performed. Keywords—Multilevel Thresholding, DEWO Optimization, Renyi entropy, Kapur entropy, Tsallis entropy. I. INTRODUCTION Thresholding is a popular technique for many image processing applications as pixels' gray level value for objects in the image and pixels' gray level value for background of the image are substantially different. Some of the common domains where thresholding can be applied are: map processing, to find lines, characters and legends[1]; quality inspection to detect defective parts of machines and to repair or remove them[2]; scene processing to capture the target[3]; medical imaging applications like cancer prediction[4], gesture classification[5] etc.; spatial and temporal segmentation of videos and other multimedia files[6]. Generally, there are more than two region of interest present in images which require more than two thresholds for efficient segmentation. Multilevel thresholding is performed on images with more than two connected components as it extracts more than two regions of interest in a simple and efficient manner. Entropy, measure of uncertainty in the information source [7], is the commonly used criterion functions that is optimized in multilevel thresholding. Entropy-based multilevel thresholding helps to achieve appropriate partition of the object image as it provides the information about the distribution of pixel levels of an image [9]. Some of the prominent entropy functions include Tsallis entropy [30], Renyi's entropy [10], Kapur's entropy [8], and Otsu's method [11]. Entropy-based multilevel thresholding using histogram is an efficient approach as it estimate threshold intensity in the image histogram through parametric and non-parametric methods. 2D histogram is a better approach than 1D histogram as it is computed by integrating pixels of original image with the filtered image obtained using local means averaging of pixels of original image. 2D histogram contains spatial information in addition to gray level distribution information and hence provide efficiency to thresholding method. Filtered image obtained through non- local means of pixels provide better spatial correlation hence is preferred over local means pixel-based filtering [12]. Further non-local means filtering calculate weighted similarity of pixels with target pixels hence provide better post filtering clarity with better results than normal 2D histogram technique that ignores information related to edges. Entropy-based multilevel image thresholding method using 2D histogram require exhaustive computation and hence is time-consuming. This drawback can be addressed by using metaheuristic algorithms, that have the capability to produce near optimal solution for all the problems that do not have any problem specific algorithm known as their solution. Metaheuristic algorithms are widely used to get optimal threshold solution with less computational cost and exhaustive exploration of the search space. Over the last few years, many researchers have shown interest in solving multilevel thresholding segmentation problem using different metaheuristic algorithms like genetic algorithm [14], ant colony optimization [13], firefly algorithm [15], differential evolution [16] etc. A meta-heuristic algorithm is said to be successful if it can achieve a proper balance between exploration and exploitation of search space. This the well-known fact that meta-heuristic algorithms perform differently for different type of problems and hence there is no ideal metaheuristic problem that exists for all problems. In this study, a new metaheuristic which is combination of Differential Evolution and Whale Optimization (DEWO) algorithms is proposed for colored image segmentation using an efficient entropy based-multilevel thresholding and 2D histogram technique. Results from three entropy