Our result of a real scene Our result of a real scene with synthetic rain (car) Our result of a real scene with synthetic rain (zooming window) Computational time and propagating video alignment (PVA) algorithm PSNR comparison (zooming window) PSNR comparison (car) Removal Accuracy comparison (car)
Rain Removal System for Dynamic Scene in Diminished Reality 〔動態場景消去實境除雨系統〕

Visibility in the rain is poor. Here we propose a rain removal system for dynamic scene in diminished reality. Our system provides a clear and comfortable view. We apply and extend chromatic prior, explore chromatic pair, dark tail removal and spatial chromatic prior. We propose an optimized intuitive alignment to solve moving camera problem. We apply anti-flicker to completely remove drizzle and defocused rain. Benefit from parallel pixel-based chromatic prior, our system has high scalability and extensibility. Our experiments reference numerous videos from prior related works, involving realistic and diverse property of rain, motion of the scene and motion of the camera. The experimental results show a satisfactory rain removal quality, especially for drizzle and defocused rain.

SUMMARY (中文總結):
下雨天可視度差。 我們提出一個動態場景消去實境除雨系統,我們的系統提供一個清晰舒適的視野。 我們使用並擴展色度先驗(chromatic prior), 發展出色度像素對(chromatic pair)、 黑尾瑕疵消除(dark tail removal)及空間域色度先驗(spatial chromatic prior)。 我們提出一個效能優化且直覺的影像對齊方法解決移動相機的問題。 我們使用反閃爍濾波器(anti-flicker)去除毛毛雨(drizzle)及散焦雨(defocused rain)。 受益於可平行化的以像素為基底的(pixel-based)色度先驗, 我們的系統具有較高的延展性及擴展性。 我們的實驗引用多個來自既有相關文獻的影片, 包含具真實性及多樣性的雨、 場景動態及相機動態。 實驗結果顯示令人滿意的除雨結果, 特別是對毛毛雨及散焦雨。

PROJECT MATERIAL: (picture gallery, video, software demo, talk slides, etc.)
  • 結果展示 (mp4/H.264)
  • 下雨場景 (mp4/H.264)