Efficient Object-based Video Inpainting

摘要

Video inpainting is the process of repairing missing regions (holes) in videos. Most automatic techniques are computationally intensive and unable to repair large holes. To tackle these challenges, a computationally-efficient algorithm that separately inpaints foreground objects and background is proposed. Using Dynamic Programming, foreground objects are holistically inpainted with object templates that minimizes a sliding-window dissimilarity cost function. Static background are inpainted by adaptive background replacement and image inpainting.