Eulerian Video Magnification 🎥🔍

A C# implementation of Eulerian Video Magnification (SIGGRAPH 2012). Reveal hidden motions and color changes in videos that your eyes can't see!

Dev Insights 💡

Motion Artifacts = Pain If the person moves their head, everything goes wild. Future plan: slap YOLO on this to detect/segment the face and stabilize the ROI. PRs welcome if you wanna help!

Why EVM is Cool This algorithm is elegant. Like, surprisingly simple but effective. It's still the fastest approach in the Video Magnification world (unless someone figures out how to make Mamba work here 👀).

The Filter Dilemma The original paper uses three kinds of filters, ideal, simple IIR, and Butterworth. For real-time processing, we went with Butterworth. We tried IIR first but it was not interpretable 🙃. Switched to Butterworth and the results got way cleaner, though there's a tiny bit of delay.

The values of r1, r2 in IIR reflect the confidence of the present and past signal. To induce a valid r1, r2, it seems to relate to r = 2*pi*Fc/Fs, and the algorithm constrains its range in [0, 1]. In the MATLAB code, the baby and wrist cases use r1=0.4 and r2=0.05, which means Fc1≈0.430/(2pi)=1.91Hz (~114 BPM) and Fc2≈0.0530/(2pi)=0.24Hz (~14 BPM).

The annoying part? You can't just say "give me 0.5-2 Hz"—you have to do math backwards to get these magic r values.

The One Cool Thing We Added

MathNet.Filtering doesn't let you configure filters like MATLAB's butter() (seriously, why??). So we wrote ButterworthHelper.cs to compute IIR coefficients directly from cutoff frequency + sampling rate.

Currently only first-order. Higher orders = PR welcome 🙏

TODO

• Implement the lambda parameter (λ) from the paper. It helps to reduce noise for Laplacian pyramid.

Quick Start

1. Open in Visual Studio
2. Click Run ▶️
3. Load a video
4. Adjust sliders
5. Watch the magic happen ✨

Requirements: .NET 8.0 + Windows

References

• Original Paper: Eulerian Video Magnification for Revealing Subtle Changes in the World (Wu et al., SIGGRAPH 2012)

2026+ Update 📅

Still interested in video-based vital signs? EVM is cool but if you're specifically doing heart rate / respiration monitoring, check out the rPPG (remote Photoplethysmography) field. Way more mature for that use case.

Recommendation: rppg-toolbox - way better accuracy for pulse detection.

---

Contributors welcome! Especially if you know signal processing better than me.