Super-resolution mosaicing from MPEG compressed video

Local object-based super-resolution mosaicing from low-resolution video

“…where M i is the ROI in the SR mosaic at the ith iteration and µ(p, K) = 1 |p| with |p| as the number of available pixels at position p. We can notice that this equation is quite similar to the SR method we presented in [27], but the restoration operator is absent. The advantage here is, if we can identify the blurring operator, the restoration is straight forward and does not require the synthesis of a restoration filter as e.g.…”

“…Here, we suppose that this is already accomplished. We used [27] for registration and [33] for moving object segmentation and tracking.…”

“…A problem in (2) and (3) is that B is defined for the SR image which is unknown. Our objective is to directly use the LR PSF instead of making an assumption on it for the SR image as we did in [27]. It can be shown that for small blurs (see Appendix A.3):…”

“…In our recent work [27], we presented an efficient SR method for the blind restoration of global blurs due to the camera motion. The main drawback of this method is that restoration is performed in frequency domain which does not allow for processing local blurs.…”

“…Based on a segmentation of moving objects, our SR method is able to estimate and locally restore blur. To this end, we extend our blur estimation method [27] for global motion blur due to the camera motion to the case of local blurs. This enhanced method can handle both, motion blur of the camera occurring in the static scene background and motion blurs of various moving objects.…”