I am using the SurfFeatureDetector to do object detection. But I find that the object's image size will affect the object detection's result. For example, if my object image is very large, it can't successfully detect, or too small can't successfully detect also. So, my question is, Is there any detector will ignore the image size? or How can I do to increase the success rate? Thanks.
Descriptors like SIFT, SURF, ORB, BRISK (MSER?), which make use of their keypoint-based scale-detection (i.e. the scale is determined during the keypoint-extraction) are actually pretty scale invariant, see http://computer-vision-talks.com/2012/08/a-battle-of-three-descriptors-surf-freak-and-brisk/ for a small evaluation. Just for clarification: this also means, the descriptor itself is typically not scale-invariant but in conjunction with its detector it gets scale-invariant.
To answer your question: typically you can adjust the scale-invariance for the detector by adjusting the number of octaves/number of pyramid layers/scale factor. If you don't want to use a scale-invariant detector then you can use a Gaussian pyramid via PyramidAdaptedFeatureDetector(), see http://docs.opencv.org/modules/features2d/doc/common_interfaces_of_feature_detectors.html#pyramidadaptedfeaturedetector.
Asked: 2013-08-02 06:35:02 -0600
Seen: 397 times
Last updated: Aug 04 '13
