| Popis: |
Super-resolution algorithms have been getting better with both computational performance and reconstruction accuracy for both single image super-resolution and video super-resolution. We propose an enhancing methodology for existing super-resolution algorithms. Our method is applicable only for videos and could speed up the computation time for any existing single image super-resolution algorithm (SISR). We consider the individual frames of a video and use the method of differences by finding a delta image for each consecutive frame after the origin frame. The non-zero values in delta frame are the only parts of that frame which are up-scaled whereas the rest of the parts are copied from the up-scaled origin image. Furthermore, the origin image is replaced with a new one after every time a scene is changed with rest of the process being the same. This algorithm could even optimize computation performance and quality, or accuracy based on the need of the viewer, that is quality could be increased or decreased at the expense of computation time. We explore up-scaling intermittent delta frames for much faster video super-resolution. Our proposed approach improves the computation time of any single image super-resolution algorithm by 40% or more for most videos when up-scaling every frame of that complete video. |
