Adaptive Wavelet Filter With Edge Compensation for Remote Sensing Image Denoising

Abstract

Advanced Geostationary Radiation Imager (AGRI) is one of the main payloads of the second-generation geostationary orbit meteorological satellite, FengYun-4A. Typically, the existence of variable stripe noise in the water vapor band remote sensing images of the AGRI greatly affects many applications, such as cloud detection, especially as one full disk image is separated into ten sub-images for transforming as soon as possible, so the denoising algorithm, which can reduce variable stripe noise and is adaptive to process using sub-images, must be built. In this paper, we propose an adaptive wavelet filter for image denoising. This approach introduces a new parameter termed weight sum variance of digital number probability (WSVODP), which is used to indicate the appropriate wavelet filter coefficients. WSVODP is only sensitive to the difference of observation targets of different sensors. Thus, our approach can learn appropriate wavelet filter coefficients fast and exactly. We built a real-world remote sensing image dataset from AGRI on FengYun-4A, and the experimental results on this dataset show that the proposed approach could effectively reduce the variable stripe noise from different observation targets. At the same time, an edge compensation method, which is fitted to the scanning model of the AGRI, is suggested to avoid ringing artifacts. Many applications, such as cloud detection with denoised images, show very good results. The proposed approach reduces the stripe noise adaptation, so the result is very steady even if the stripe noise varies with different targets, and edge compensation ensures that there are no obvious ringing artifacts in the full disk image joined by the ten sub-images.