Despeckling of SAR Images using BEMD based Adaptive Filters

Abstract

2023 newlineThe Synthetic Aperture Radar (SAR) images were heavily affected by speckle newlinenoise, because SAR systems generates the images through the scattered indicators. newlineThe patterns of speckle noise are not consistent, whose appearance changed based on newlinescattering and reflection properties of light. Therefore, many despeckling approaches newlinewere developed to remove the speckle noise from SAR images. However, the newlineconventional methods were resulted in reduced subjective and objective performance. newlineThus, this research work is focused on implementation of Bidimensional Empirical newlineMode Decomposition (BEMD) based adaptive filters for despeckling. newlineThe SAR images are decomposed into multiple bands such as Bidimensional newlineIntrinsic Mode Functions (BIMF) using BEMD. Here, the BIMF bands holds the newlinehigh-frequency and low frequency components with texture regions of SAR. Further, newlinethe primary BIMF band is filtered by various adaptive filters such as Lee, Kuan, Frost newlineand GMAP. Then, all the BIMF bands are added with adaptive filter outcome, which newlinegenerates the despeckled SAR image. The simulations are conducted on the SAR newlineimages obtained from Sandia National Laboratory. The Matlab R2020a software with newlineIntel Core i7 G50 on a laptop equipped with 2.40GHz CPU and 4GB RAM was used to newlinecarry out all simulations for implementation of various filters. Different window sizes newlinelike 3_3, 5_5, and 7_7 have been used for simulations along with speckle variances newline0.01 and 0.05. The simulation results have shown good quality reconstruction in the newlineSAR image, and the proposed filtering method able to suppress the speckle noise in the newlinefiltered image. Finally, the proposed BEMD based GMAP filter resulted in superior newlinesubjective and objective performance. Here, the GMAP filter resulted in low time newlinecomplexity i.e., 24.48 seconds, which is less as compared to Lee, Kuan and Frost filters. newline