Stripe Noise Elimination with a Novel Trend Repair Method on the Push-Broom Thermal Image

Stripe noise on the remote sensing image is a general phenomenon that not only degrades the image quality but also severely limits its application. While the classical statistical method is effective in correcting common stripes caused by inaccurate calibration of relative gains and offsets between detectors, it falls short in correcting other nonlinear stripe noises resulting from minor nonlinear changes or random contaminations that occur within the same detector. Therefore, this paper proposes a novel trend repair method based on adjacent normal columns to rectify the trend of the defective column by considering the geospatial structure of the contaminated pixels to remove residual stripe noises after histogram matching. The proposed trend repair method is compared with the piece-wise method using the GF5-02 VIMI (Visual and Infrared Multispectral Imager) thermal Band 9 image to evaluate its effectiveness. Streaking (streaking metrics), SSIM (structural similarity), and PSNR (peak signal-to-noise ratio) are employed to assess the performance of the new method. Experimental results indicate that the proposed trend repair method removes residual stripe noises effectively after histogram matching.