Abstract
We propose a video reconstruction method that can generate a high-resolution and high-frame–rate video from a lower pixel-rate video. To achieve this, the proposed method uses color-separated videos with different spatio-temporal resolutions. Reconstruction process consists of two steps. First, low-resolution videos are up-sampled by using local correlation coefficient between different color videos. Second, a low-frame rate video is temporally interpolated using motion between frames. To reduce residual errors in motion estimation, the method simultaneously estimates both reconstructed videos and “reconstructed motions” which are derived from reconstructed high-resolution videos. Experimental results show that residual errors in both motion and reconstructed videos are significantly reduced.
