Chaotic fitness-dependent quasi-reflected Aquila optimizer for superpixel based white blood cell segmentation

The crisp partitional clustering techniques like K-Means (KM) are an efficient image segmentation algorithm. However, the foremost concern with crisp partitional clustering techniques is local optima trapping. In addition to that, the general crisp partitional clustering techniques exploit all pixels in the image, thus escalating the computational time. In order to prevail over local trapping problem as well as balance the escalating computational time, this paper presents a Chaotic Fitness-Dependent Quasi-Reflected Aquila Optimizer (CFDQRAO) based crisp clustering strategy which is an improved variant of one of the Nature-Inspired Optimization Algorithms (NIOA), i.e., Aquila Optimizer (AO). The chaotic fitness-dependent quasi-reflection based Opposition Based Learning (OBL) has been incorporated into classical AO to make it a more competent optimizer. Alternatively, Simple Linear Iterative Clustering (SLIC)-based super-pixel images have been explored as input to the clustering technique to lower the computational time of the suggested clustering strategy. In this research, the author provides the results of an experiment performed using images of blood pathology for the purpose of segmenting white blood cells (WBCs). The results reveal the preeminence of the proposed CFDQRAO technique over other tested NIOAs in regard to the optimization ability and consistency. Further, the proposed SLIC-CFDQRAO clustering strategy proved itself better than other SLIC-NIOA based clustering strategies and even SLIC-KM in terms of visual analysis and the values of segmentation quality parameters.

Automated summary

This paper presents an improved crisp clustering strategy based on the Aquila Optimizer combined with chaotic fitness-dependent quasi-reflection and Simple Linear Iterative Clustering (SLIC)-based super-pixel images. Experimental results using blood pathology images for segmenting white blood cells (WBCs) show that the proposed CFDQRAO technique outperforms other tested NIOAs in terms of optimization ability and consistency. Additionally, the proposed SLIC-CFDQRAO clustering strategy is better than other SLIC-NIOA based strategies and even SLIC-KM in terms of visual analysis and segmentation quality parameters.