A Multilevel Image Thresholding Using Particle Swarm Optimization Debashis Mishra #1 , Isita Bose #2 , Utpal Chandra De *3 , Bishwojyoti Pradhan #4 # School of Computer Engineering, KIIT University Bhubaneswar, Odisha, India 1 debashis.engg@gmail.com 2 isitabose89@gmail.com 4 bishwojyoti2@gmail.com * School of Computer Application, KIIT University Bhubaneswar, Odisha, India 3 Abstract— Image Thresholding is one simplest method of image segmentation, which partitions the image into several objects on the basis of one or more threshold values. Threshold values are the values chosen from the intensity values of the image. In this paper, 8-bit unsigned gray scale images are taken as sample where the intensity values ranges from 0 to 255. Here Kapur's entropy criterion method is used which i a function of threshold values and is optimized by the advanced swarm based optimization technique named as particle swarm optimization (PSO). Particle swarm optimization is a nature-inspired methodology which mimics the food searching technique of birds. In this paper PSO takes Kapur's entropy criterion method as fitness function and gives the optimized threshold values to segment the image. This method gives the better result using small swarm size and few number of iterations comparing to the traditional image thresholding technique.. utpal@kiit.ac.in Keyword- Image Segmentation, Thresholds, Multi-variable Thresholding, PSO, Fitness function, Gray- scale image. I. INTRODUCTION Image Processing is one of the emerging research area, which generally manipulates digital raw image for better understanding of image. It is used for various purposes such as noise reduction from image[7], image enhancement, image retrieval, image understanding, image segmentation etc. Image segmentation is a process of partitioning an image into several objects for better understanding of the image. It is helpful to detect a particular object from the image which is mostly used in medical science as to detect tumour from ultra sound image, in space science to detect space objects or to locate any specific region. This is also used in biometric applications viz. finger print recognition, iris detection etc. Swarm Intelligence (SI) is one sub-domain of soft computing methodologies which is a population based optimization technique. It generally mimics the social behaviours of different animals from nature to solve the problems. In today's era of solving quadratic equations, Swarm Intelligence (SI) helps to do optimization with optimum accuracy and in less time. Particle Swarm Optimization(PSO)[5], is a kind of swarm intelligence which mimics the social behaviour of birds while searching foods. In this paper, basic PSO technique is adopted to optimize two threshold values using extended Kapur's entropy criterion method [4] as fitness function. In this paper the basic idea about image thresholding technique is covered in section II. In section III, the basic model for particle swarm optimization is illustrated clearly with an algorithm. Section IV of this paper focuses on the proposed work. Experiments and conclusions are explained in section V and VI respectively. II. IMAGE THRESHOLDING As already discussed in section 1 of this paper, it is somehow clear that image thresholding is one type of image segmentation technique which segments or partitions one image into several object to detect different regions of same colour or gray level intensities. It is the simplest way to segment an image using threshold values. Threshold values are the intensity values chosen from the colour (for colour image) or gray level (for gray-scale image) intensities of the image. Generally threshold values are chosen from the up and down of the image histogram which is shown in figure 1. On the basis of choosing these threshold values and segmenting image, image thesholding can be classified into different categories as described below. But in this paper our only focus is on multi-variable thresholding. Debashis Mishra et al. / International Journal of Engineering and Technology (IJET) ISSN : 0975-4024 Vol 6 No 2 Apr-May 2014 1204