3

My question is about the meaning of Hu's seven invariant moments.

As far as I know, some moments have a meaning; i.e. the zeroth order refers to the area of the image. Also as stated in http://mathworld.wolfram.com/topics/Moments.html a lot of features can be extracted from different order of moments. But the question is what features are extracted by Hu moments? I know that the key point of his work is that these features are invariant to the RTS (rotation, translation, and scale). But my question is what are these features describing? What is the relationship between these features and the object in the image?

Thanks for your help.

Omid
  • 31
  • 1
  • 4

1 Answers1

4

The Hu moments, proposed in 1962 in the article “Visual pattern recognition by moment invariants”, have no direct correspondence to physical aspects of the shape from which they are extracted. Rather, they were mathematically formulated to be invariant under translation, scale and rotation.

Alceu Costa
  • 9,733
  • 19
  • 65
  • 83
  • Thanks for your reply Alceu about the invariant moments. Besides that, I'm also interested to use the “moments” in fingerprint identification. Do you know what are the best moments or order of moments to use for fingerprints? Of course any set of moments will give me similar values for different impression of the same finger and dissimilar values for different impressions of different fingers, but I don't want to just use these moments while I don't know what are they describing; I'm after a meaningful set of moments. – Omid Jul 24 '12 at 05:36
  • I believe that image moments (and most image descriptors) are not very adequate for fingerprint recogntion. Usually, fingerprint recognition methods are based on "minutia features" (http://en.wikipedia.org/wiki/Fingerprint_recognition#Minutia_features). – Alceu Costa Jul 24 '12 at 12:12