5
$\begingroup$

I have large database of pictures, about a few million.

I need an algorithm to find pictures which are similar to those in the database, up to a scale factor or a crop and shift (I have only these types of variations).

I know about histogram matching and perceptual hash, but is there a faster algorithm that exist for these purposes?

$\endgroup$
  • 2
    $\begingroup$ You're talking about a complete research field, namely Content Based Image Retrieval! Start googling on that for some algorithms :) $\endgroup$ – Geerten Aug 17 '12 at 8:49
  • $\begingroup$ @Geerten I know about CBIR, but I have restricted task.It seems it called 'image copy detection' or 'image duplicate detection' $\endgroup$ – mrgloom Aug 17 '12 at 9:20
  • $\begingroup$ The way you described it in your question sounds like CBIR. Maybe you can expand your question a bit more, and provide some more detail about the application? $\endgroup$ – Geerten Aug 17 '12 at 9:35
  • $\begingroup$ @Geerten It's like CBIR, because it's part of CBIR.CBIR is wider task with variable "similarity metric", but at beginning I need just duplicate detection with taking into account only scale,crop,shift. $\endgroup$ – mrgloom Aug 17 '12 at 9:41
  • $\begingroup$ Although not about images, but music, I would recommend reading A highly robust audio fingerprinting system to get an idea about the architecture behind this type of copy detection systems. I also think CBIR based on (SIFT) features is promising to deal with cropping. $\endgroup$ – Maurits Aug 17 '12 at 11:20
1
$\begingroup$

Take a look at "Features for Image Retrieval" text.

$\endgroup$
1
$\begingroup$

For this you can use local features like SIFT, SURF, FREAK, etc.

$\endgroup$

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.