Hybrid denoising method for space optical remote sensing images based on compressed sensing
Abstract
Spaceborne optical remote sensing images are inevitably degraded by two dominant noise components including Gaussian noise and Salt-and-Pepper noise during acquisition and transmission, while massive high-resolution data imposes stringent burdens on onboard transmission bandwidth and storage. To address these challenges, this study employs a hybrid denoising framework that synergizes Compressed Sensing (CS) with spatial-domain filtering techniques. The proposed CS based hybrid strategy utilizes a learned dictionary and an improved Orthogonal Matching Pursuit (OMP) algorithm to achieve high-precision reconstruction in low-dimensional sub-spaces. By utilizing differentiated post-processing, this method overcomes the fidelity-smoothness trade-off prevalent in conventional filters. A systematic quantitative comparison is conducted against classical mean filtering and median filtering methods. The comparative analysis confirms that the hybrid method maintains robust restoration across diverse noise types, effectively reconciling the demands of noise suppression, detail preservation, and data compression under resource constraints.