Performance evaluation and analysis of population-based metaheuristics for denoising of biomedical images

Introduction: Medical imaging is an ever sensitive domain, where the results are capable enough to affect the lives of patients. The corruption of images with noise is an obstacle faced in the domain of medical imaging and processing. The images can be corrupted by different types of noises which have to be approached differently for denoising. The main idea behind denoising is to remove the noise content in the images while preserving the edges. Denoising while preserving the edges is a challenge to the processing models, since it is not able to distinguish the edges from the noise components, both being high frequency and similar in characteristics. The necessity of an optimization algorithm arises, to optimize the denoising process of images subjective to the property of each image. Objective: The objective of this study was to analyze various optimization algorithms and find how they work on biomedical image denoising problems. Methods: The optimization algorithms are used in a framework involving FIR filtering and using PSNR as an objective function for the problem. The results upon denoising the images using the different optimization algorithms are then compared visually and numerically to decide upon the algorithms best suited for the task. Results: Coyote Optimization Algorithm (COA) is seen to have a clear advantage over other algorithms in denoising the biomedical images in the best way. However, analyzing the convergence plots yields a different perspective showing that COA does not converge as fast as other algorithms like Artificial Bee Colony Algorithm. Conclusion: The study opens up the various niches for improvement and development in the optimization algorithms in the domain of biomedical image denoising. It concluded COA to have the best results under the given conditions of FIR filtering and PSNR objective function. Further, the convergence of COA can be improved, and the algorithms which show better convergence can be used in combination with other objective functions to yield better results.