I recently was studying gabor filters and came to know that they have a shortcoming that they do have a minimum but some reasonable dc component.
now my question is why is it necessary to remove the dc component (using log gabor).
If there are any other limitation of gabor filters then please do tell me.


gabor filter formed a sine/cosine wave with a gaussian, where the center of frequency of the filter is specified by the frequency of sine-cosine and the band width of the filter is specified by the width of the gaussian (valid perfectly inside a certain range and outside it, attenuates the frequency). In other words, more than one gabor filter to cover the spectrum. the gabor filter can be designed for a band-width of 1 octave maximum.

you can find more details about gabor and log-gabor

  • $\begingroup$ Yes I figured it out myself...But thanks for your reply :) $\endgroup$ – noobmaster69 Feb 11 '14 at 10:38
  • $\begingroup$ @adil: Just curious, does the log-gabor filter perform better than the gabor filter? $\endgroup$ – meta_warrior Feb 26 '14 at 2:34
  • $\begingroup$ in my case I got 4-5% accuracy increment while testing when I replaced gabor filter's with log gabor ones in the feature extraction process.... $\endgroup$ – noobmaster69 Feb 26 '14 at 6:07
  • $\begingroup$ Do you know the reason for the increment in accuracy? @adil $\endgroup$ – meta_warrior Mar 24 '14 at 14:56
  • $\begingroup$ It is written there that natural images are better coded by gaussian function on log scale(as in log-gabor) rather than normal scale(as in gabor). So the features are more accurate. Also images with wider spectrum cant be filtered using one gabor filter because of octave limitation but it is not the case of log gabor... Its all there :) $\endgroup$ – noobmaster69 Mar 25 '14 at 16:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.