Multi-frame super resolution using frame selection and multiple fusion for 250 million pixel images

In this paper, we propose a multi-frame super-resolution (SR) method using a frame selection and a multiple fusion for 250 million pixel image processing. When a tiny part (with a target that is several kilometers away from the camera) is enlarged, quality of the part tends to get poor due to spatial deformations and noise mainly caused by long distance and atmosphere turbulence. In this paper, we first propose an adaptive frame selection method that select a small part of frames with small blur according to the corresponding edge images detected by sobel filters. Then, we propose a multiple fusion scheme to reconstruct selected images in order to removing deformation. In our multiple fusion scheme, all of the alternate images are reference images for reconstruction, by this way both deformation and noise can be removed effectively without expensive computation. Our proposed method for enhancing the quality of 250 million pixel images performs better than conventional multi-frame super-resolution methods in accuracy, simplicity and ease of implementation so that it is quite suitable for commercial use.