Speckle noise removal in SAR images using Multi-Objective PSO (MOPSO) algorithm

Abstract

SAR images are inherently affected by speckle noise, and although attempts made earlier to remove speckle succeeded, there is still the challenge of preserving the edges of images. This is due to the smoothing effect of most of the earlier algorithms that work on thresholding coefficients in the transform domain. There exists a trade-off between denoising and the ability to preserve edges in selecting a suitable threshold. Estimation of an optimal threshold is a major concern and is compounded by the requirement for concurrent smoothing of noise and preservation of structural/edge information in an image. Considering the search for an optimal threshold as exhaustive and the requirements as contradictory, we model this as a Multi-Objective Particle Swarm Optimization (MOPSO) task and propose a MOPSO framework for despeckling an SAR image using a Dual-Tree Complex Wavelet Transform (DTCWT) in the frequency domain. Two counteractive reference metrics, such as Peak Signal-to-Noise Ratio (PSNR) and Mean Structural Similarity Index Metric (MSSIM), and non-reference metrics such as the alpha-beta (αβ) and Despeckling Evaluation Index (DEI) have been used as the objective functions of MOPSO. An optimal threshold derived from this multi-objective optimization is chosen for despeckling the SAR images. The proposed solution has been found to outperform state-of-the-art filters such as Lee, Kaun, Frost and SAR-BM3D filters. Also, the proposed MOPSO framework superior than the competing optimization technique Multi-Objective Evolutionary Algorithm (MOEA) based on Differential Evolution (DE) framework for despeckling.