Sift Implementation :: Keypoints from various octaves

Question

I am trying to implement SIFT in MATLAB.

I got the key points for (say) 4 different octaves by localizing the local maxima and minima in the DOG (Gaussian difference) space. However, I am at a loss from what to do with key points from the last 3 octaves.

Do I include them at key points for the original image, if so, how do I translate from a reduced image to the original, for example, a pixel (i, j) in a 256 x 256 image (2 octaves) per pixel (i ', j') in the image 512x512 (1 octave). I tried to sift through a lot of textbooks, but did not find anything convincing.

Answer 1

It is not clear what you mean by "last 3 octaves"? About translation - you multiply the resulting scale (from blur kerkel) and (x, y) twice for the 2nd octave, by 4 for the third octave, etc.