A low-complexity, motion-robust, spatio-temporally adaptive video de-noiser with in-loop noise estimation

Noise in video influences the bit-rate and visual quality of video encoders and can significantly alter the effectiveness of video processing algorithms. Recent advances offer high quality de-noising at a fairly high computational complexity by increasing the spatio-temporal support and evaluating intelligent weights for combining these samples to remove noise while preserving the signal. The lower complexity methods, typically, either over-blur the video or introduce motion artifacts and temporal flicker. By reusing the motion vectors generated by a video encoder, a low incremental complexity de-noiser is proposed in this paper that is capable of achieving a high level of noise reduction and signal preservation with a reduced spatio-temporal support. In addition, the approach lends itself to dynamically estimating the noise variance used for controlling the level of filtering. The proposed approach performs on par with the spatial non-local means de-noising algorithm for stationary background sequences and can be improved for motion sequences with motion-compensated temporal filtering.