Differential exponential entropy-based multilevel threshold selection methodology for colour satellite images using equilibrium-cuckoo search optimizer

Recently, the entropic based multilevel threshold selection methods use 2D histogram, which is constructed using the local averages, leading to a loss of edges. Further, the computation of the entropy using the diagonal pixel values only leads to a loss of information. Nevertheless, traditional 2D histogram based multilevel thresholding methods suffer from efficiently retaining the spatial correlation information. In addition, the conventional entropy uses logarithmic function, which has inherent problems, thereby, reducing the accuracy at some situations. To solve these issues, a differential exponential entropy (DEE) -based multilevel threshold selection methodology is proposed. To suppress the high magnitude peaks in the 2D histogram, the normalized local variance is used while the construction. A novel objective function is suggested to compute the DEE. A new Equilibrium-Cuckoo Search Optimizer (ECSO) is suggested to maximize the DEE. For testing, standard benchmark functions are used. The results are compared with the physics-based Equilibrium Optimizer (EO) and the nature-inspired Cuckoo Search Algorithm (CSA). Different benchmark colour satellite images are acquired from the Landsat Image Gallery database for the experiment. The performances are compared with the state-of-the-art methods. Different metrics such as PSNR, SSIM and FSIM are used for the image quality assessment. A statistical analysis is presented in terms of the Box plots. Our proposed DEE-ECSO outperforms the other techniques. The suggested algorithm would be useful for segmentation of the brain MR images for biomedical engineering applications.

Automated summary

This paper proposes a multilevel threshold selection method based on differential exponential entropy (DEE) to address the challenges of edge loss, insufficient retention of spatial correlation information, and accuracy reduction due to logarithmic function in traditional approaches. The method uses normalized local variance in histogram construction to suppress high magnitude peaks, and introduces a new objective function and optimizer to maximize DEE. Experimental results show that the proposed method outperforms other techniques in terms of image quality assessment metrics.