Timeline for Scale and Rotation invariant feature descriptors
Current License: CC BY-SA 3.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 7, 2012 at 22:16 | history | edited | Andrey Rubshtein | CC BY-SA 3.0 |
added 513 characters in body
|
| Nov 3, 2012 at 22:41 | comment | added | penelope | Looking forward to it :) | |
| Nov 3, 2012 at 21:24 | comment | added | Andrey Rubshtein | @penelope, I guess now that I was not clear enough. I will try to expand it tomorrow. Thanks you for your comments. | |
| Nov 3, 2012 at 20:05 | comment | added | penelope | Sorry, I don't understand what you mean by "a thumbnail of the patch" can you explain? And the descriptor: normalized corss-correlation of you patch is what makes it rotationally invariant, not the fact that your patch was a ring or circle. | |
| Nov 3, 2012 at 19:53 | history | edited | Andrey Rubshtein | CC BY-SA 3.0 |
Imroved answer
|
| Nov 3, 2012 at 19:44 | comment | added | Andrey Rubshtein | @penelope, why not? Consider the following descriptor - a thumbnail of the patch, and you compare it with other descriptors with normalized cross-correlation. You can't mistake it with other objects in this way. | |
| Nov 3, 2012 at 16:24 | comment | added | penelope | we are talking about keypoint (feature) descriptors. If we were talking about features (interest points), then detecting circular patches might be useful -- they are rotationaly invariant in combination with every descriptor. But a descriptor calculated at a circular patch is not - a white circle with horizontal diameter in black, and with a vertical one would produce very different descriptors if rotationally invariant method isn't used | |
| Nov 2, 2012 at 19:01 | history | answered | Andrey Rubshtein | CC BY-SA 3.0 |