Revision history [back]

This is a very wide ranged question.

FREAK is essentially a descriptor. A descriptor creates a unique representation of an input feature, by applying some image processing around a set of selected interest points.

This description can be used to train basicly every kind of approach you want. First approach could be to train a SVM detector. Another approach could be to use a randomn forest approach if you want to detect multiple classes.

So basically, it is a trial and error, comparing detectors, and see which one works best in your application. Possible detectors can be found here :

• http://docs.opencv.org/modules/features2d/doc/features2d.html
• http://docs.opencv.org/modules/objdetect/doc/objdetect.html

Basically it depends what you want to do. Comparing against the exact same stored model, would suggest using keypoint matchers and use the amount of found matches to define the quality. This is called object recognition.

Or are you actually searching for more general object detectors. Then the second link could give you a better clue.

Some more input could be usefull to give a better answer.