The thermal infrared image has lower signal-to-noise ratio (SNR) in a scene with small signal dynamic range.In many real applications,the scene is towards one continuous monotonous direction.Under this consideration,a scene-based video denoising method was proposed.By using low rank matrix approximation (LRMA) theory,a prior knowledge was introduced to construct strict observations.The denoising low-rank matrix of observations was solved via weighted nuclear norm minimization (WNNM) and then reconstructed video signal.Simulation showed that the method had high peak signal-to-noise ratio (PSNR) and denoising robustness.Processed data of ocean remote sensing verified the practical effect of the method.The method has certain application value in a field of remote sensing and rescue.