Video Filtering Using Separable Four-Dimensional Nonlocal Spatiotemporal Transforms

The large number of practical application involving digital videos has motivated a significant interest in restoration or enhancement solutions to improve the visual quality under the presence of noise. We propose a powerful video denoising algorithm that exploits temporal and spatial redundancy characterizing natural video sequences to reduce the effects of noise. The algorithm implements the paradigm of nonlocal grouping and collaborative filtering, where a four-dimensional transform- domain representation is leveraged to enforce sparsity and thus regularize the data. Moreover we present an extension of our algorithm that can be effectively used as a deblocking and deringing filter to reduce the artifacts introduced by most of the popular video compression techniques. Our algorithm, termed V-BM4D, at first constructs three-dimensional volumes, by tracking blocks along trajectories defined by the motion vectors, and then groups together mutually similar volumes by stacking them along an additional fourth dimension. Each group is transformed through a decorrelating four-dimensional separable transform, and then it is collaboratively filtered by coeffcients shrinkage. The effectiveness of shrinkage is due to the sparse representation of the transformed group. Sparsity is achieved because of different type of correlation among the groups: local correlation along the two dimensions of the blocks, temporal correlation along the motion trajectories, and nonlocal spatial correlation along the fourth dimension. As a conclusive step, the different estimates of the filtered groups are adaptively aggregated and subsequently returned to their original position, to produce a final estimate of the original video. The proposed filtering procedure leads to excellent results in both objective and subjective visual quality, since in the restored video sequences the effect of the noise or of the compression artifacts is noticeably reduced, while the significant features are preserved. As demonstrated by experimental results, V-BM4D outperforms the state of the art in video denoising. /Kir1