Video denoising based on transform domain minimum mean square error

Denoising is one of the most common and important tasks in video processing systems and abundant efforts have been made on video denoising nowadays. In this paper, we propose a novel denoising scheme based on minimum mean square error (MMSE) filter in the 2D transform domain, which we call 2DTD-MMSE. The current input noisy frame is processed block-by-block, and for every block, the current noisy observation and multiple prediction blocks found by motion estimation (ME) in denoised previous frames as well as noisy future frames constitute a 2D observed representation array. Afterwards, 2D transform is applied to every block in the representation array, and every transform coefficient of current block is estimated by weighted average of the coefficients in the same frequency position of all the transformed blocks. The weighting coefficients are adaptively determined through MMSE for every block after the estimation of statistical parameters in the transform domain. Experimental results on commonly used test sequences demonstrate that the proposed 2DTD-MMSE achieves comparable or favorable performance when compared to several state-of-the-art algorithms.