A Weighted Guided Filtering-Based Multidomain Fusion Destriping Method

Abstract

An infrared camera is affected by the photon effect, temperature changes, time drift, and other factors when operating in orbit, which makes the ground nonuniformity correction coefficient invalid, resulting in nonuniformity stripes in the infrared images and restricting their practicality in further analysis and applications. The existing destriping methods often suffer from the loss of image details and artifact generation. To solve this problem, we proposed a weighted guided filtering-based multidomain fusion destriping approach that leverages the structural, directional, and spectral characteristics of stripe noise. First, we addressed the issue of artifacts caused by Fourier-domain filtering through an adaptive filtering approach that employs a variable threshold to minimize filtering-induced artifacts and obtain clearer guided images. Furthermore, capitalizing on the directional properties of wavelet decomposition effectively separates image information from stripe information. To integrate the advantages of both approaches, we employed a weighted guided filter to seamlessly fuse the guided image with the wavelet decomposition image. In terms of quantitative metrics, the proposed method generally beats the other five comparative methods, with significant improvements in image PSNR, SSIM, NIQE, and mean relative deviation (MRD), particularly for complex images where the enhancements were more pronounced. These experimental results collectively demonstrate the significant progress achieved by the proposed method in effectively reducing stripe noise, better preserving the original structural details of the image, and suppressing the occurrence of artifacts.

Keywords

• Fourier transform
• image fusion
• stripe noise
• thermal infrared
• wavelet decomposition
• weighted guided filtering (WGF)