OpenCV/C++ - SIFT keypoint detection and corresponding descriptors

Question

I have an image. I used Canny edge detector,and then I apply Sift on it.

Here the result :

I don't understand the result. Why have i some interest points on some black areas and not a lot where I expect the more ? Besides, what does the circle size refer to ?

I took a look on the internet but it's not totally clear and I don't find any information about it.

Can anyone explain?

Answer 1

SIFT stands for "Scale-invariant feature transform". It detects keypoints on multiple scales. The circles on your image probably refer to the scale on which the keypoint was detected.

the keypoint positions in black areas don't seem completely unlikely to me: if you suppose that the area they look at is about 2 or 3 times the size of the circle, than the keypoints in the black zone make a lot more sense: it does not just detect a black zone but a zone like this for example; which is distinctive:

important remark a keypoint does not mean that only that exact point is different from the rest; it means that that point, together with its surroundings is particular. In the case of SIFT it means that SIFT will be able to detect the same point (or zone if you will) even if you scale the image. The keypoint location returned is then the "centre" of that zone if you will.