Comparison of Point Feature Detectors and Descriptors

Comparison of Point Feature

Detectors and Descriptors

in the context of cultural heritage

Acknowledgement:

The research project has received funding from the People Programme (Marie Curie Actions) of the European Union's Seventh Framework Programme FP7-PEOPLE 2007-2013 under REA grant agreement ITN2013 no 608013.

References

• Alahi, A., Ortiz, R., & Vandergheynst, P. (2012). Freak: Fast retina keypoint. CVPR 2012, pp. 510–517.

• Bay, H., Tuytelaars, T., & Van Gool, L. (2006). Surf:

Speeded up robust features. ECCV 2006, pp. 404–417. • Lowe, D. G. (1999). Object recognition from local

scale-invariant features. ICCV 1999, 2, pp. 1150–1157. • Lowe, D.G. (2004). Distinctive image features from

scale-invariant keypoints. IJCV 2004, 60(2), pp. 91–110. • Rublee, E., Rabaud, V., Konolige, K., & Bradski, G.

(2011). Orb: an efficient alternative to sift or surf. ICCV

2011, pp. 2564–2571.

Stathopoulou E., Stavropoulou G., Georgopoulos A., Van Gool L., Ioannides M.

No. Detected Points

SIFT

FREAK

SURF

Descriptor Extractors

Feature extraction is the basis of many computer vision applications, ranging from SfM to classification and retrieval methods. Local features have been proven to work best for such applications. When it comes to cultural heritage, computer vision methods are often used to solve problems of 3D reconstruction and repository management. However, there is not a one-method-fits-all solution for feature extraction and the generation of suitable feature sets should be tailored according to the needs of each application. Along this line of thought, we compare several state-of-the art 2D detectors and descriptor. Case study: Asinou church, Cyprus.

SIFT

SURF

FAST

Keypoint Detection

Number of Keypoints

ORB

Good Matches

The selection of detectors and descriptors is application depended. Our tests were performed on highly textured images with presence of repetitive patterns. FAST outperforms the other two feature detectors in time efficiency as well as in the number of detected keypoints. Regarding the descriptors, ORB is faster and thus more suitable for real-time computer vision applications, but results many false positives. On the other hand, SIFT is more robust for the detection of “good” matches, while SURF appears more balanced between efficiency and time. Future work will include experiments on more datasets with different kind of transformations between image pairs.

0 5 10 15 20 25

Computational

Time

SIFT

SURF

ORB

FREAK

Computational Time

SIFT

SURF

FAST