Microsoft Research Community

Blending artifacts

rated by 0 users
This post has 7 Replies | 2 Followers

Top 150 Contributor
Posts 6
pavel Posted: 11-27-2008 2:29 PM

Hello,
did you see the blending artifacts like this?
Kind regards,
Pavel 

Blending artifacts

Top 10 Contributor
Posts 139

I would expect the blending technique in ICE to do better than this.  What size image set is this?  If it is a small image set I could take a look at this particular case and debug a bit.

Matt Uyttendaele
Top 150 Contributor
Posts 6

Hi Matt,
the demo set was about 60 images (15columns x 4rows), each has 2048x1768 resolution. The overap was about 20%. The presented area is about 500-1000 pixels long. But similar transitions were visible in more areas. The total size of all images is about almost 700MB resulting in approximately 24k x 6k resolution.
I tried to prepare a smaller set with just 2-8 images, but I got smooth transitions only.
I have to check the public availability of such images due to my company restrictions.
It seems like being connected to so large number of images - probably. Maybe it is related to some limitations mentioned by you at http://community.research.microsoft.com/forums/t/2204.aspx
I did not notice such artifacts in some other examples with more structures without so smooth transitions.
Let me know whether this information is sufficient for you or how do you think to continue.
Kind regards,
Pavel

Top 10 Contributor
Posts 139

Interesting.  You are the second person to report this - that the transitions look good on small sets but not on larger sets.  We must have a bug lurking in there - I'll have to investigate a bit.

Thanks for taking the time to report this.

Matt Uyttendaele
Not Ranked
Posts 1

I have a set of 5 images, all hand-held, and ICE does a terrible job stitching them. There are those artefact lines all over, and they don't only separate different exposures, but even different parts of the image. The horizon is teared apart and quite a bit lower at the other side of this gap. Overall, there doesn't seem to be any blending done, the images are just geometrically corrected and then put together at more or less suitable places. If the two images don't perfectly fit together, you can see that artefact line going through the entire resulting image.

I've used autostitch before, but in this set, it constantly ignores 2 of 4/5 images. I'm now trying with hugin...

Top 10 Contributor
Posts 139

I expect that you are giving ICE a set that is difficult to register - perhaps there is a lot of parallax?.  If you want to post the originals or result somewhere we can take a look. 

I think that overall people have been generally happy with the blending e.g. http://www.flickr.com/photos/venndiagram/3676219888/ 

Matt Uyttendaele
Top 10 Contributor
Posts 48

I most commonly see blending artefacts on open water where waves/ripples will differ from one frame to the next. Sometimes these are blended well, other times there's a very obvious zig-zag line.

Top 10 Contributor
Posts 139

I agree. Our blending technique does not work particulalry well for water.  We simple perform a brighten/darken for each image so that the seams between them disappear.  This works really well for things like exposure, white-point, and small amounts of vignetting.  However not so well for water.  An enblend like technique may be preferable for the water. 

Matt Uyttendaele
Page 1 of 1 (8 items) | RSS
©2009 Microsoft Corporation. All rights reserved. Terms of Use | Trademarks | Privacy Statement | Feedback