B.K. Panigrahi et al. (Eds.): SEMCCO 2012, LNCS 7677, pp. 17–24, 2012.
© Springer-Verlag Berlin Heidelberg 2012
Multilevel Image Thresholding Based on Tsallis Entropy
and Differential Evolution
Soham Sarkar
1
, Swagatam Das
2
, and Sheli Sinha Chaudhuri
3
1
Electronics and Communication Engineering Department, RCC Institute of Information
Technology, Kolkata – 700015, India
2
Electronics and Communication Sciences Unit, Indian Statistical Institute,
Kolkata – 700108, India
3
Electronics and Telecommunication Engineering Department, Jadavpur
University, Kolkata – 700032, India
sarkar.soham@gmail.com, swagatamdas19@yahoo.co.in,
shelism@rediffmail.com
Abstract. Image segmentation is known as one of the most critical task in
image processing and pattern recognition in contemporary time, for this purpose
Multi Level Thresholding based approach has been an acclaimed way out.
Endeavor of this paper is to focus on obtaining the optimal threshold points by
using Tsallis Entropy. In this paper, we have incorporated a Differential
Evolution (DE) based technique to acquire optimal threshold values.
Furthermore, results are compared with two state-of-art algorithms- a. Particle
Swarm Optimization (PSO), and b. Genetic Algorithm (GA). Several image
quality assessment indices are applied for the performance analysis of the
outcome derived by applying the proposed algorithm.
Keywords: Multilevel Image Segmentation, Tsallis Entropy, Differential
Evolution, MSSIM, WPSNR.
1 Introduction
IMAGE segmentation, the process of discriminating objects from its background in
pixel level, has become the utmost component of image analysis. Over the years
segmentation is being applied as a basic step for several computer vision applications
like feature extraction, identification, image registration etc. Image segmentation done
via bi-level thresholding that subdivides the image into two homogenous regions,
based on texture, histogram, edge etc., uses only one threshold value.
In the year 2004, bi-level maximum Tsallis entropy (MTE) based image
segmentation was proposed by Portes de Albuquerque et al [1]. Proposed technique of
image segmentation is based on work done by Tsallis et. al. [2], which is an extended
version of Havrda and Charva’t paper in 1967 [3]. Later multi-level image
segmentation gained popularity for its ability for sub-dividing the image into more
than one segment. It makes the image more useful for the later analysis and studies.
However, the computation complexity of these methods had increased to a significant