Video denoising

Опубликовано 6 сентября 2016, 17:46

It is well known that noise can be introduced into video during capturing, storage and transmission. It is especially bad in dark regions in the video where the signal-to-noise ratio is low. Such noise is typically visually disturbing and can cause serious problem to many video coding systems such as MPEG-1/2/4, and H.261/3/4. This is because many precious bits would need to be spent to encode this undesirable noise in the video. As MPEG-2, MPEG-4, H.263 and H.264 are used widely in consumer electronics such as DVD, Blu-ray, DVB-C/T/S/H, DMB, HDTV, PDA, IPTV, video conferencing and streaming, there is a great need for video denoising. In this talk, we will introduce a class of optimal linear filter in temporal domain to suppress noise. Even when the filter length is very small, the denoising results can be very significant. In pre-filtering applications, these filters require robust motion estimation (in noisy condition) which requires much computation. We will show a class of fast motion estimation especially designed for denoising purpose. We will also introduce a formulation that can integrate the denoising filter seamlessly into the video encoders and decoders such as MPEG-1/2/4 and H.261/3/4. The resulting complexity is greatly reduced and the video coding performance greatly improved.