Adaptive wavelet restoration of noisy video sequences

In this paper, we present a novel algorithm for restoration of noisy video sequences. A video sequence is first transformed into an optimal 3D wavelet domain using basis functions adapted to the contents of the sequence. Assuming that all the major spatiotemporal frequency phenomena present in the sequence produce high amplitude transform coefficients, a modified form of the BayesShrink thresholding method is used to suppress the noise. In order to reduce the effects of Gibbs phenomenon in the restored sequence, translation dependence is removed by averaging the restored instances of the shifted sequence. The algorithm yields promising results in terms of both objective and subjective quality of the restored sequence.