0

I am trying to patch one image that has missing data with another taken from another angle. I had some success with computing features i.e. SIFT and finding a homography with surfaces with unique features. I later do blending to make the images nice.

The trouble appears when I am trying to patch images of the floor which can be featureless or have many non-unique features. Repeating this method has very poor results as far too many features are computed and not well matched. Is there a much better way to do this? I was looking at plane estimation algorithms. At worst if I could estimate the plane and very roughly align the images that would be fine.

It would be great to hear a different view point on this problem. Here are 2 example images I have tried (a very hard case I know).

Image to patch

2nd Photo

Edit: My camera lens right now is faulty. The images would be more clear than this normally so I am aware this is an extra difficult case.

Thomas1982
  • 1
  • 3
  • is this basically the same as `image mosaicing` or `panorama imaging`? Maybe you find working approaches with those keywords. – Micka Dec 17 '15 at 15:43
  • Similar yes. The camera can be a fair bit more offset than in mosaicing. I am also using images of different resolutions and quality. – Thomas1982 Dec 21 '15 at 16:11
  • should be no problem. But I only got experience with Sift/SURF on image mosaicing, so I might be no help (since you've tried Sift already). – Micka Dec 21 '15 at 16:13

0 Answers0