I am having a difficulty in understanding the parameters orientationNormalized and scaleNormalized in the FREAK descriptor. Any idea on what they mean or do?
OpenCV FREAK doc : http://docs.opencv.org/modules/features2d/doc/feature_detection_and_description.html#freak-freak
On
As old-ufo said: with "orientationNormalized" the orientation of the keypoint will be estimated and stored in the keyPoint's parameter 'angle'. Which can than be used for better matching.
The scaleNormalization indeed influences the border-check, but the background is that if you turn on scaleNormalization, the size of the pattern (used to extract the descriptor) will be scaled according to this formula. (having in mind, that the 'minimum' keypoint size is by default 7)
To better understand the scaling also be sure to look into the buildPattern() method!
(and to read the according paper)
"orientationNormalized" means that it estimates keypoint orientation, which is good when orientation invariance is needed.
Scale normalization, as far as I understood from the source code, influences only check if the whole keypoint is lies inside image.