In this paper, we present a method of generating an all-focused photography using an omni-directional camera. With a convex mirror and a single projective camera, we are able to take 360 degree panorama picture. However picture taken is often not in focus because of the depth variation for each pixel. By taking several pictures of the same scene with different focal planes, we classify well-focused/ill-focused pixels, comparing the suggested metric for blur estimation, so that a patch at a pixel of interest is the most focused. By empirically testing all different parameters of the system and the results from them, we analyze the pros and cons of each parameter and the presented idea. Depending on sensitivity of the estimation, the approximated depth might be somewhat unreliable. Hence, we are currently extending the system to infer the depth of a scene as well.
|저 자||Dong-Gun Lee, Da-Hyun Jeon, Jean-Charles Bazin, In So Kweon|
|학 회||26th ICROS Annual Conference (학부생 포스터 세션)|