Author Affiliations
1School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou, Gansu 730070, China2Gansu Provincial Engineering Research Center for Artificial Intelligence and Graphics & Image Processing, Lanzhou, Gansu 730070, Chinashow less
Fig. 1. AR recognition registration process based on text features
Fig. 2. FAST feature point detection
Fig. 3. FREAK retina sampling mode
Fig. 4. Center symmetrical sampling point pair
Fig. 5. Improved corner points of FASText
Fig. 6. Matching results of text images of class A. (a) SURF algorithm; (b) ORB algorithm;(c) FREAK algorithm; (d) improved FREAK algorithm
Fig. 7. Matching results of text images of class B. (a) SURF algorithm; (b) ORB algorithm;(c) FREAK algorithm; (d) improved FREAK algorithm
Fig. 8. Matching results of text images of class C. (a) SURF algorithm; (b) ORB algorithm;(c) FREAK algorithm; (d) improved FREAK algorithm
Fig. 9. Recognition registration for image A. (a) Positive registration; (b) distance and rotation change registration; (c) occlusion registration;(d) perspective change registration
Fig. 10. Recognition registration for image B. (a) Positive registration; (b) distance and rotation change registration; (c) occlusion registration;(d) perspective change registration
Fig. 11. Recognition registration for image C. (a) Positive registration; (b) distance and rotation change registration; (c) occlusion registration; (d) perspective change registration
Parameter | SURF | ORB | FREAK | Improved FREAK |
---|
A | B | C | A | B | C | A | B | C | A | B | C |
---|
Average running time /ms | 36.6 | 55.2 | 57.6 | 34.5 | 38.2 | 39.3 | 33.8 | 36.5 | 37.6 | 31.8 | 32.7 | 33.1 | Contrast of averagerunning time /% | -13.1 | -40.7 | -42.5 | -7.8 | -14.3 | -15.7 | -5.9 | -10.4 | -11.9 | - | - | - | Average number offeature point pairs | 358 | 572 | 607 | 336 | 509 | 535 | 324 | 469 | 498 | 302 | 338 | 342 | Contrast of average numberof feature point pairs /% | -15.6 | -40.9 | -43.6 | -10.1 | -33.5 | -36.0 | -6.7 | -27.9 | -31.3 | - | - | - | Precision /% | 69.6 | 64.2 | 62.3 | 76.9 | 75.5 | 74.6 | 78.7 | 76.2 | 75.4 | 87.8 | 85.4 | 84.6 |
|
Table 1. Comparison of average running time, average number of feature point pairs, and matching precision of algorithms
Algorithm | Imageprocessing | Featureextraction | Featurematching | Poseestimation | Modelrendering | Total time |
---|
SURF | 1.3 | 21.6 | 18.1 | 8.7 | 6.4 | 56.1 | ORB | 1.3 | 18.2 | 17.5 | 9.8 | 6.4 | 53.2 | FREAK | 1.3 | 16.8 | 15.4 | 9.8 | 6.4 | 49.7 | Improved FREAK | 1.3 | 15.1 | 13.8 | 9.8 | 6.4 | 46.4 |
|
Table 2. Average processing time of different algorithmsms