Multi-Frequency Analysis of Fully Polarimetric Speckle Filter based on Morphology Using Visual Quality Indices & Support Vector Machine

Abstract

In this study the multifrequency data is utilized to test the performance of a novel fully polarimetric speckle filter based on Morphological operations. The main objective of this work is to determine the filter performance by inspecting the visual quality of the image using visual quality indices followed by a classification study on PolSAR images. PolSAR images are acquired from the P, L, & C bands of the Microwave range. The most commonly present noise in Polarimetric Synthetic Aperture Radar (PolSAR) images is the Speckle Noise. The speckle noise is removed using a novel fully polarimetric filter based on morphological operations. After Multi-looking the PolSAR images are converted into the Coherency Matrix commonly known as the T-Matrix. The Morphological filter is applied on the elements of the T-Matrix. Two opening and closing operations are considered while applying the Morphological filter to see the effect of multiple closing and opening operations. The opening and closing are based on the principles of erosion and dilation. The filtered images are then tested for visual quality and classification accuracies. The visual quality of the images after processing with morphological operations are carried out using the full reference & no reference quality metrics. The full reference quality metrics considered are Mean Square Error (MSE), Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM) and Multi-Scale Structural Similarity Index (MS-SSIM). The no reference quality metrics considered are Blind/Reference less Image Spatial Quality Evaluator (BRISQUE), Natural Image Quality Evaluator (NIQE), and Perception based Image Quality Evaluator (PIQE). The commonly used edge preservation index for SAR images is also calculated. The classification accuracy is determined using the Support Vector Machine (SVM), with Radial Basis Function (RBF) as the kernel, along with cross validation. The visual quality of the image and the classification accuracies are compared with the existing filters at all the P, L, & C band frequencies. It is observed that the proposed technique can reduce the speckle significantly, maintain visual quality and give good classification accuracies, at all the three frequencies..