Motion-Compensated Wavelet Video Denoising

We study a low complexity, motion-compensated recursive video denoising scheme in the wavelet domain. To preserve weak spatial textures, both spatial and temporal filtering are carried out in the wavelet domain. We justify the proposed approach by evaluating the relative increase of mean squared error (MSE) with respect to the optimal estimator. Various multiresolution motion estimation schemes are studied to allow us to exploit the high temporal correlation present in most video. Kalman filtering is then applied to the wavelet coefficients along motion trajectories to efficiently suppress noise. We show experimentally that our wavelet-based recursive denoising compares favorably with other wavelet-based denoising approaches. Specifically, we can preserve both strong and weak spatial details while removing noise.