image diff
5 views
Skip to first unread message
Dr. Sanith Wijesinghe
Oct 14, 2022, 10:11:09 PM10/14/22
to ope...@googlegroups.com
Hi there,
I was wondering if openMPF comes with any packages that allow you to perform an image diff?
The two images submitted for comparison would be a picture of a property from the ground-level and then a separate image of the property from a satellite image. These would be properties that are damaged, and in some cases there will be no property -i.e. just rubble.
Any suggestions on tools/techniques would be very much appreciated.
Thanks!
Dr. Sanith Wijesinghe
Technical Fellow
MITRE Labs
571.419.7036 (m)
Jeffrey Robble
Oct 14, 2022, 10:44:25 PM10/14/22
to OpenMPF
Hi Dr. Wijesinghe,
Sorry to say, that is not a use case we've considered for OpenMPF.
One option would be to perform some kind of image-based 3D reconstruction. That is, build up a 3D model of the property of interest before the damage, and compare that to a 3D model of the property after the damage. It often requires multiple images from different perspectives to build up one 3D model. See https://stackoverflow.com/a/10870643. Once you have two 3D models that you can align on top of each other, you would need to perform some kind of 3D object comparison.
Usually the problem people are trying to solve is image matching. Generally, the solution is SIFT: https://www.analyticsvidhya.com/blog/2019/10/detailed-guide-powerful-sift-technique-image-matching-python/. The problem here is that if the images are truly different, because one shows a property with damage, and one without, then SIFT won't be able to select as many keypoints for comparison, but maybe that's what you want. I assume there is some way to use SIFT to generate a similarity metric which you can threshold on. If similarity is low, maybe there's property damage. Relevant paper: https://www.cs.ubc.ca/~lowe/papers/ijcv04.pdf. It would not be difficult to write an OpenMPF component that uses SIFT since it's readily available in OpenCV.
Sorry to say, that is not a use case we've considered for OpenMPF.
One option would be to perform some kind of image-based 3D reconstruction. That is, build up a 3D model of the property of interest before the damage, and compare that to a 3D model of the property after the damage. It often requires multiple images from different perspectives to build up one 3D model. See https://stackoverflow.com/a/10870643. Once you have two 3D models that you can align on top of each other, you would need to perform some kind of 3D object comparison.
Usually the problem people are trying to solve is image matching. Generally, the solution is SIFT: https://www.analyticsvidhya.com/blog/2019/10/detailed-guide-powerful-sift-technique-image-matching-python/. The problem here is that if the images are truly different, because one shows a property with damage, and one without, then SIFT won't be able to select as many keypoints for comparison, but maybe that's what you want. I assume there is some way to use SIFT to generate a similarity metric which you can threshold on. If similarity is low, maybe there's property damage. Relevant paper: https://www.cs.ubc.ca/~lowe/papers/ijcv04.pdf. It would not be difficult to write an OpenMPF component that uses SIFT since it's readily available in OpenCV.
- Jeff
Reply all
Reply to author
Forward
0 new messages