时空自适应的分焦平面偏振视频PCA去噪

PCA-based spatial-temporal adaptive denoising of DoFP video for microgrid polarimeters

分焦平面式(DoFP)偏振成像探测器通过集成式微偏振阵列实现偏振信息的实时获取。然而由于成像过程中存在噪声,对后续的偏振图像去马赛克超分辨、场景偏振信息解算产生了严重影响。基于主成分分析(PCA)提出一种时空自适应DoFP视频数据去噪算法,对于每个待去噪的DoFP图像块,在其局部时空邻域内选取相似的图像块,然后利用主成分分析对其去噪。该算法充分利用DoFP视频数据的时空信息构建训练样本,且块匹配过程无需采用运动估计,可直接用于DoFP视频数据去噪。进一步提出基于双边滤波的残余噪声去除算法,从而得到更好的去噪效果。通过模拟与真实数据对所提算法进行实验验证,结果证明:所提算法可有效抑制噪声,在相同测试条件下,所提算法优于现有算法。

Division of focal plane(DoFP) polarization imaging detector are composed of integrated micro-polarizer array on a focal plane array sensor, which make the DoFP polarimeters capture the polarization information real-time. However, it is difficult to perform the DoFP demosaicking and reconstruct the polarization information due to noise. A PCA-based spatial-temporal adaptive denoising method was presented to work directly on the DoFP videos. For each DoFP patch to be denoised, similar patches were selected within a local spatial-temporal neighborhood. The principal component analysis was performed on the selected patch to remove the noise. The spatial-temporal information of DoFP video was used to construct the sample patches. The proposed method worked directly on the DoFP video without explicit motion estimation. And then a fast bilateral filtering algorithm was used to remove the residual noise in different polarization channels of DoFP images. The experimental results on simulated and real noisy DoFP sequences demonstrate that the proposed denoising method can significantly reduce the noise-caused polarization artifacts and outperform other denoising methods.