Adaptive restoration of multispectral datasets used for SVM classification

Removing noise from images while keeping its important details unchanged is a challenging issue in image restoration. In this paper, we propose a novel approach based on partial differential equations (PDE) in order to mitigate three well-known types of noises from remote sensing data while important features such as edges are preserved. In the presented method, after performing the Watershed-based segmentation as a preprocessing step, optimum values of PDE parameters are adaptively found based on the noise type and the image texture. In order to evaluate the performance of the proposed algorithm, Peak Signal-to-Noise Ratio (PSNR) criterion is applied. Moreover, feeding the original/noisy/denoised images into SVM classifier and exploring the classification ratios are suggested as an application-based assessment. The gained results prove a considerable enhancement both in quantitative metrics (PSNR and MSE) and SVM classification ratios (from 71.71% to 95.07%).