Temporally coherent video editing using an edit propagation matrix

摘要

A frame-by-frame edit propagation system for video is introduced in this study. The proposed approach propagates the editing parameters by multiplying a large and dense matrix, which we refer to as the Edit Propagation Matrix, to a vector that consists of the editing parameters of the pixels in the preceding frame. The proposed approach has two main advantages. First the computational cost depends only on the size of each frame and is independent of the total size of the input video. Second, this approach only needs to consider the relationship between two successive frames. This approach can be applied to a wide variety of computer graphics applications such as color manipulation, stylization, colorization, and tonal stabilization.