Eulerian video magnification
Eulerian Video Magnification — Wu, Rubinstein et al., MIT CSAIL, SIGGRAPH 2012
Definition
Eulerian video magnification is a computer-vision technique that amplifies imperceptibly small motions and color variations in a video — for example, the subtle skin-color pulse caused by blood flow, or the millimeter-scale motion of a guitar string. It was introduced by MIT CSAIL in SIGGRAPH 2012 and reframed earlier Lagrangian motion-magnification methods into a per-pixel temporal-filtering pipeline.
Verbatim from the source
“The first Eulerian method to amplify small motions and color variations in videos.”
Technical detail
The Eulerian framing is the conceptual breakthrough: rather than tracking individual moving features (the Lagrangian approach, which is computationally expensive and brittle), the algorithm decomposes the video into spatial frequency bands using a Laplacian pyramid, applies a temporal bandpass filter at each pixel within each band, scales the filtered result, and adds it back to the original. The result is a video where motion in a chosen frequency band is amplified.
Limitations: works best for small motions (large motions get distorted by the linear approximation), and noise in the video gets amplified along with the signal. Phase-based motion magnification (Wadhwa et al. 2013) addresses both limitations and is the basis for later real-time iOS implementations.
Where it appears
- Slick Engineering Research Used here
Slick Engineering's Motion Magnification case study reverse-engineers Eulerian and phase-based motion magnification into real-time GPU-accelerated iOS implementations using custom Apple Metal compute shaders.