feature_detection_and_description.rst



Feature Detection and Description

Note

An example explaining keypoint detection and description can be found at opencv_source_code/samples/cpp/descriptor_extractor_matcher.cpp


FAST
Detects corners using the FAST algorithm
Detects corners using the FAST algorithm by [Rosten06].


[Rosten06]


Rosten. Machine Learning for High-speed Corner Detection, 2006.


MSER
Maximally stable extremal region extractor.
class MSER : public CvMSERParams
{
public:
    // default constructor
    MSER();
    // constructor that initializes all the algorithm parameters
    MSER( int _delta, int _min_area, int _max_area,
          float _max_variation, float _min_diversity,
          int _max_evolution, double _area_threshold,
          double _min_margin, int _edge_blur_size );
    // runs the extractor on the specified image; returns the MSERs,
    // each encoded as a contour (vector<Point>, see findContours)
    // the optional mask marks the area where MSERs are searched for
    void operator()( const Mat& image, vector<vector<Point> >& msers, const Mat& mask ) const;
};

The class encapsulates all the parameters of the MSER extraction algorithm (see
http://en.wikipedia.org/wiki/Maximally_stable_extremal_regions). Also see http://code.opencv.org/projects/opencv/wiki/MSER for useful comments and parameters description.

Note

(Python) A complete example showing the use of the MSER detector can be found at opencv_source_code/samples/python2/mser.py


ORB
Class implementing the ORB (oriented BRIEF) keypoint detector and descriptor extractor, described in [RRKB11]. The algorithm uses FAST in pyramids to detect stable keypoints, selects the strongest features using FAST or Harris response, finds their orientation using first-order moments and computes the descriptors using BRIEF (where the coordinates of random point pairs (or k-tuples) are rotated according to the measured orientation).


[RRKB11]
Ethan Rublee, Vincent Rabaud, Kurt Konolige, Gary R. Bradski: ORB: An efficient alternative to SIFT or SURF. ICCV 2011: 2564-2571.


ORB::ORB
The ORB constructor

ORB::operator()
Finds keypoints in an image and computes their descriptors

BRISK
Class implementing the BRISK keypoint detector and descriptor extractor, described in [LCS11].


[LCS11]
Stefan Leutenegger, Margarita Chli and Roland Siegwart: BRISK: Binary Robust Invariant Scalable Keypoints. ICCV 2011: 2548-2555.


BRISK::BRISK
The BRISK constructor

BRISK::BRISK
The BRISK constructor for a custom pattern

BRISK::operator()
Finds keypoints in an image and computes their descriptors

FREAK
Class implementing the FREAK (Fast Retina Keypoint) keypoint descriptor, described in [AOV12]. The algorithm propose a novel keypoint descriptor inspired by the human visual system and more precisely the retina, coined Fast Retina Key- point (FREAK). A cascade of binary strings is computed by efficiently comparing image intensities over a retinal sampling pattern. FREAKs are in general faster to compute with lower memory load and also more robust than SIFT, SURF or BRISK. They are competitive alternatives to existing keypoints in particular for embedded applications.


[AOV12]


Alahi, R. Ortiz, and P. Vandergheynst. FREAK: Fast Retina Keypoint. In IEEE Conference on Computer Vision and Pattern Recognition, 2012. CVPR 2012 Open Source Award Winner.


Note

An example on how to use the FREAK descriptor can be found at opencv_source_code/samples/cpp/freak_demo.cpp


FREAK::FREAK
The FREAK constructor

FREAK::selectPairs
Select the 512 best description pair indexes from an input (grayscale) image set. FREAK is available with a set of pairs learned off-line. Researchers can run a training process to learn their own set of pair. For more details read section 4.2 in: A. Alahi, R. Ortiz, and P. Vandergheynst. FREAK: Fast Retina Keypoint. In IEEE Conference on Computer Vision and Pattern Recognition, 2012.
We notice that for keypoint matching applications, image content has little effect on the selected pairs unless very specific what does matter is the detector type (blobs, corners,...) and the options used (scale/rotation invariance,...). Reduce corrThresh if not enough pairs are selected (43 points --> 903 possible pairs)