Here's where things stand. Getting sufficient resolution from the digitized 16mm film to see the full rivet and deformation pattern on the patch is "pushing the envelope." Nothing like this has ever been done. Not surprisingly, problems arise. By combining sequential frames with super-resolution technology, Jeff was able get a much clearer image but, in a film taken with a wide-angle lens in excellent lighting where everything should be in focus, he could see individual lines of rivets on the vertical tail but not on the fuselage. He eventually determined that, for some reason, possibly a flaw in the camera's aperture gate, the tail was in focus but the fuselage was not. Using Artificial Intelligence he was able to shift the focus of the frames to the fuselage. Once that was accomplished he had a much clearer picture of the patch. By combining just 2 frames he could see the lines of large rivets along the bottom and aft edges of the patch. By combining more frames with super-resolution augmented with AI he was able to see much more detail - but still not enough to make out the lines of small rivets on the fuselage and patch. Something else was wrong and it took a long to time figure out what. It turned out that the assumption that the best results could be obtained by combining sequential frames was incorrect. Some frames are good. Some are not so good. It now appears that the best results are obtained by using AI to select the best frames in the entire group and then combining those frames with super-resolution. Jeff has done a proof-of-concept experiment with a limited number of selected frames. The next step is to try that technique with a much larger number of frames. Ideally, he'll get to the point of being able to see the lines of small rivets on both the fuselage and the patch. Once we're sure we're able to see all there is to see we'll be able to tell whether the rivet pattern on the patch matches the rivet pattern on 2-2-V-1.
Author
Topic: Status of super-resolution of video? (Read 10810 times)
