Motion blur resistant method for temporal video denoising

In this paper, we propose a fast temporal video filter which aims to minimize blur and blocking artifacts, and handle signal-dependent noise. In order to overcome motion blur problems in block-based temporal filters, an accurate motion detection has been developed using two levels of reliability. At the first level, we use temporal data blocks to coarsely estimate local motion error and noise. Then at a finer level, averaging weights are calculated utilizing fast convolution operations. Utilizing regional noise re-estimation and the noise level function (in the case that it is known), our method is designed to adapt to signal-dependent noise and noise overestimation. Motion vectors are estimated through a fast hybrid motion estimation method which combines two conventional methods, compounding the strengths of each for a more efficient estimation. The proposed method is easy to implement and results show it rivals state-of-the-art methods in both quality and speed.