Whale Optimization Algorithm for Color Image Segmentation using Supra-Extensive Entropy

Abstract

Image segmentation plays an important role for image analysis. Image thresholding technique is one of the most effective segmentation techniques. Although, bi-level thresholding is widely applied to segment non-complex color images, however, bi-level thresholding is not suitable in case of color complex images. In case of color complex images which contain multiple objects, only multi-level thresholding works efficiently. The conventional thresholding approaches give efficient results for bi-level thresholding, but the time complexity of the conventional approaches may be excessively high for color image multilevel thresholding due to search multiple threshold values for three (red-green-blue, RGB) components. Thus, color image multilevel thresholding segmentation can be considered as NP-hard combinatorial optimization problem because the time complexity of the searching procedure increases exponentially as levels of thresholding increase. Here, the major objective is to search optimal threshold values for segmenting the color image into appropriate segments. In this paper, Supra-Extensive entropy based new objective function is designed to find optimal threshold values for segmenting the color image into multiple segments. For optimizing the proposed objective function, two well-established population based optimization approaches Whale Optimization Algorithm (WOA) is explored. Such approach is called WOA-based SEEMT. The proposed approach is compared with Grey Wolf Optimizer (GWO) called GWO-based SEEMT algorithm. Experiments are performed on six color benchmark images in terms of optimal threshold values, peak signal to noise ratio (PSNR), uniformity, structure similarity (SSIM) index, mean structure similarity (MSSIM) index, number of iterations and CPU time. The experimental results show that the there is no significant difference between the performance of WOA-SEEMT and GWO-SEEMT algorithms in terms quality parameters PSNR, uniformity, SSIM index and MSSIM index while WOA-SEEMT algorithm is much faster than GWO-SEEMT from computational point of view.