This paper presents a new “marine snow” removal method for deepsea videos based on robust temporal-spatial decomposition. For deepsea videos, the contents of adjacent frames are almost identical or change very little except for the rapidly moving “marine snow” as well as noise, indicating that there exists high temporal-spatial correlation between the successive frames. Based on this observation, we first robustly approximate the deepsea video to recover its background using online robust principal component analysis in a sub-video-by-sub-video manner. Since the structure information of background cannot be well preserved during the background modeling, we further extract such information from the approximation error to compensate the obtained background, which is also formulated as a constrained convex optimization problem. The experimental results demonstrate that our proposed method can achieve comparable or even better results than the state of the art approach.
- Deepsea videos
- Image enhancement
- Robust principal component analysis
ASJC Scopus subject areas
- Media Technology
- Hardware and Architecture
- Computer Networks and Communications