A Spatiotemporal Edge-preserving De-noising Method for High-quality Video

A novel spatiotemporal method for de-noising high-quality video is introduced. We also give a method for fast and precise estimation of the moving part of the frame based on quad-tree decomposition of the difference image of the adjacent frames and tri-color block matching. This method is applied to the sequence before the main processing, dividing the set of neighboring frames into variable size blocks with known motion vectors. Temporal de-noising is performed by averaging the motion-compensated frames weighted by the matching quality for each block. Thus, the better matches contribute more to the result. The edges and parts that are not matched in any of the neighboring frames are replaced with the original image and its de-speckled version respectively. Excellent performance of the method is shown by its application to part of high-definition sample footage