Spatio-Temporal Video Denoising by Block-Based Motion detection

This paper proposes a new video denoising technique where spatially adaptive noise filtering in wavelet (transform) domain is combined with temporal filtering in signal domain. AWGN is being considered which behaves as Gaussian random variable. In this paper, spatial filtering of individual frames is done in the wavelet domain, and the filtering between the frames is done by recursive temporal filter. Spatial filtering is done by taking wavelet transform of individual frames and then modifying the wavelet coefficients by spatially adaptive bayesian wavelet shrinkage method. The denoising artifacts and residual noise differ from frame to frame which produces unpleasant visual effect. Hence filtering in time domain is essential. Temporal filtering is based on a simple block based motion detector and on selective recursive time averaging of frames. This technique outperforms sequential spatio-temporal filters, 2-D spatial filters and 3-D (spatio-temporal) in terms of visual quality as well as quantitative (PSNR) performance measures.