Spectral-spatial 3D dynamic trimmed median filter for removal of impulse noise in remotely sensed images

Multiband images get immense importance in the field of remote sensing image analysis and applications. The remote sensing images get affected by unwanted impulses because of the sensor and bit errors during acquisition and transmission, which hampers remote sensing image analysis and applications by altering the image scenes. The restoration of corrupt satellite images gain extensive consequences and is indeed a great challenge for remote sensing applications. Several spectral filters are proposed in the literature for the last few decades but they fail to preserve the valuable circumstances. In this study, a spectral-spatial 3D dynamic median filter with spatial-spectral property is proposed to deal with the impulse noise in the remotely sensed images without compromising the image quality. Our objective is to remove the impulse noise and restore the original details of the corrupted images with greater efficiency by constructing a 3D dynamic kernel, which can adaptively change the number of neighbours of the corrupted pixel, depending upon the degree of distortions in the neighbourhood. We have also concentrated on the high correlation among the consecutive spectral channels for restoring the information of the corrupted pixel. The proposed filtering method has been rigorously analyzed for varying features of satellite image datasets in the presence of different noise densities. The outcome of the proposed filter is compared with the existing state-of-the-art and recent non-linear filtering methods of image denoising and restoration. The performances and efficiency of the proposed filter have been shown through analysing its superiority over the considered and most accepted advanced filters.