Ask Your Question

SIFT Descriptor content

asked 2018-01-06 13:22:13 -0500

sama_z gravatar image

I got really confused by the content of SIFT descriptor. As I got the descriptor should contain the orientation of keypoints. For each keypoint 128 numbers different orientations. And then a feature vector which is histogram of these orientations could be generated. My questions are first after saving the descriptor file I found that the orientations are not more than 150. I saved image with its keypoints using drawKeypoints function and DRAW_RICH_KEYPOINTS as the flag and in documentation it is said that : "For each keypoint the circle around keypoint with keypoint size and orientation will be drawn." In that image I could see some keypoints' orientations look like 300!! Why this is happened? and second the orientations should be in 8 dimensions, is that right? It means I should have just these numbers: 45,90,135, 180, 225, 270, 315 and 360. But I have more than these 8 directions in descriptor! Almost any number between 0 and 150. Why? Does it mean to make a histogram of keypoints I should group all orientations to those 8 groups myself?

Please someone give me an answer, There is a while I am reading about it and still have these issues to understand SIFT completely. Many thanks in advance

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2018-01-08 03:41:15 -0500

The documentation discusses the binning quite clearly, right here

edit flag offensive delete link more



Thanks @StevenPuttemans , but the link did not explain that much. I have read more papers and I found that I misunderstood, Actually the descriptor does not consist orientations, but vectors of features. So, my question was wrong from the base!

sama_z gravatar imagesama_z ( 2018-01-09 05:23:33 -0500 )edit
Login/Signup to Answer

Question Tools

1 follower


Asked: 2018-01-06 13:22:13 -0500

Seen: 120 times

Last updated: Jan 06 '18