Characterization of correlated noise in video sequences and its applications to noise removal
Nemanja Petrović, Vladimir Zlokolica, Ljubomir Jovanov, Bart Goossens, Aleksandra Pižurica, Wilfried R. Philips · Ghent University Academic Bibliography (Ghent University) · 2007
Video sequences in TV and surveillance systems usually contain noise which decreases the visual quality and the performance of various post-processing tasks in the video chain.Usually only white Gaussian noise is assumed within these video application.However in practice that assumption does not always hold and results in poor denoising performance of standard video enhancement algorithm.In order to solve these problems we propose a new adaptive wavelet-based video denoising method.The method consist of a novel noise modeling scheme and the proposed noise-adaptive spatio-temporal filter.Specifically, the correlated (granulated) noise is characterized by a covariance matrix estimated from the noisy video sequence.Based on the estimated noise covariance and the presence of signal in a spatially local ares, the shrinkage factor for each wavelet band and the spatial position is determined, for spatial denoising.The spatial filtering is followed by the recursive temporal filtering in order to remove the remainder of noise.