Destriping high-resolution satellite imagery by improved moment matching

High-resolution satellite imagery is frequently contaminated with vertical and non-periodic stripes, because of the differential sensitivities of the detector elements to incoming radiation, and the imaging mechanism of the pushbroom-type instruments. The high spatial resolution also implies more structural details and limitations in the spectral range and the number of bands. In this article, focusing on these factors, an improved moment-matching method is proposed for the automatic destriping of high-resolution satellite imagery. Two novel approaches - Gaussian mixture model based grey slicing and an adaptive local window - are specifically designed to facilitate the homogeneity of the local image. In this way, the limitation of moment matching is eliminated. The stripes are suppressed and serious artefacts are avoided, while the radiation characteristics and structural details are preserved. The performance of the proposed approach was evaluated and compared both visually and quantitatively, employing Ziyuan-1-02C of China (ZY-1-02C) and Shijian-9A of China (SJ-9A) panchromatic images. The experimental results confirm the superiority of the proposed approach in suppressing stripes while preventing unwanted artefacts and structural distortion.