TIEVisionFeatureAlgorithm

Declaration

TIEVisionFeatureAlgorithm = (
  ievFeatureORB,
  ievFeatureSIFT,
  ievFeatureAKAZE,
  ievFeatureKAZE,
  ievFeatureBRISK
  );

Description

The keypoint detector and descriptor used by TIEVisionFeaturesDetector.

Value	Description
ievFeatureORB	Oriented FAST and Rotated BRIEF. Very fast, but finds fewer correct matches
ievFeatureSIFT	Scale-Invariant Feature Transform. The most reliable matching, especially with changes of scale and viewpoint, but slower
ievFeatureAKAZE	Accelerated KAZE. A good balance of accuracy and speed
ievFeatureKAZE	Uses a nonlinear scale space. Accurate keypoint positions, but the slowest
ievFeatureBRISK	Binary Robust Invariant Scalable Keypoints. Fast

Comparison

Results of aligning the three image pairs of the ImageAlignment demo (Customs Form, Book and Certificate) with detect, findHomography and warpPerspective, using the default image size limits:

Algorithm	Aligned all 3 pairs (maxFeatures = 2000)	Aligned all 3 pairs (maxFeatures = 500)	Matches (Customs / Book / Certificate)	Accuracy	Time	Peak Memory
ievFeatureORB	Yes	Yes, but less accurate on Book	300 / 300 / 300 *	Good, weakest on Certificate	60 ms	74 MB
ievFeatureSIFT	Yes	Yes	209 / 575 / 280	Best or equal best on all 3	300 ms	308 MB
ievFeatureAKAZE	Yes	Yes	188 / 202 / 72	Good	550 ms	248 MB
ievFeatureKAZE	Yes	Yes	258 / 75 / 103	Good	1400 ms	370 MB
ievFeatureBRISK	Yes	No, failed on Certificate	156 / 128 / 90	Good	400 ms	120 MB

* ORB keeps a fixed percentage of its matches (goodMatchPercentage), so its count does not indicate quality.
Time is per pair. Peak memory is for a 2.4 MP image (whole process). Results will vary with the images, so if alignment fails or is inaccurate, try another algorithm.

Recommendation

◼ievFeatureSIFT (the default) for the most reliable results
◼ievFeatureORB when speed or memory matters most, e.g. video or batch processing

TIEVisionFeatureAlgorithm