Search code, repositories, users, issues, pull requests...
This tool processes videos in Blacmagic RAW format and applies a temporal lowpass filter to significantly reduce noise for videos recorded in low light environments. Essentially this results in a virtual "very long exposure time" for video.
A digital image sensor records images with noise. This noise, like most random processes in nature, is roughly Gaussian distributed. Statistics tells us that computing the mean of a large number of Gaussian distributed samples eventually results in the true mean of the underlying Gaussian distribution.
When you use a long exposure time on a standard photo camera, something very similar happens internally: light is collected over an extended period of time to significantly reduce the image sensor noise.
This tool computes a moving average of successive video frames and therefore mostly reduces the random image sensor noise. Unfortunately this has two drawbacks:
However, the noise reduced images are "good enough" that applying extra gain becomes feasible to improve brightness significantly beyond what's reasonable without this noise reduction in place.
