-1

Which one is the feature in SIFT .Is key Point or descriptor .I am trying to cluster .which one should I consider for clustering and why?. This cluster data is used for classification.

madan ram
  • 1,260
  • 2
  • 19
  • 26

1 Answers1

1

The descriptor is what you mean by "feature". Keep in mind that a SIFT descriptor is made of 128 int elements, which will make your classifying task really hard (from a computational point of view) unless using some kind of dimensional reduction

morepaolo
  • 631
  • 3
  • 9
  • The classification of SIFTs is not hard, "bag of visual words" model uses it and it is common technique. – old-ufo Feb 27 '14 at 07:24
  • 1
    Bag of visual words, in its original version, is NP-hard. – morepaolo Feb 27 '14 at 07:32
  • 1
    Keypoint is a vector of four elements representing location (x,y), scale and orientation of your descriptor. The descriptor itself is a vector of 128 elements describing "histograms of local gradients", in a way which is invariabt to scale and rotation, and robust against noise and illumination. Lowe's paper can be a bit difficult to read without some math prerequirements, but it can be really enlightening – morepaolo Feb 27 '14 at 07:51
  • how clustering done on this 128 int element.I have set of image – madan ram Feb 27 '14 at 07:51
  • Try reading "bag of visual words" as suggested on top... In fact, how to cluster your data depends from the usage they're intended for... – morepaolo Feb 27 '14 at 08:00