Applying image processing techniques independently to each frame of a video often leads to temporal inconsistencies. We propose a gradient-domain approach to recover temporal consistency by inferring the temporal regularity from the original unprocessed video. We characterize the frequency properties and stabilize a wide range of image processing techniques.

Nicolas Bonneel, LIRIS-CNRS
James Tompkin, Harvard School of Engineering and Applied Sciences
Kalyan Sunkavalli, Adobe Research
Sylvain Paris, Adobe Research
Deqing Sun, Harvard School of Engineering and Applied Sciences
Hanspeter Pfister, Harvard School of Engineering and Applied Sciences