1 / 17

Triangle-Constraint for Finding More Good Features Piero Zamperoni Best Student Paper Award, ICPR 2010

Triangle-Constraint for Finding More Good Features Piero Zamperoni Best Student Paper Award, ICPR 2010. Xiaojie Guo and Xiaochun Cao Computer Vision Laboratory Tianjin University, China. Tianjin University Computer Vision Lab .

howell
Télécharger la présentation

Triangle-Constraint for Finding More Good Features Piero Zamperoni Best Student Paper Award, ICPR 2010

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Triangle-Constraint for Finding More Good Features Piero Zamperoni Best Student Paper Award, ICPR 2010 XiaojieGuo and Xiaochun Cao Computer Vision Laboratory Tianjin University, China Tianjin University Computer Vision Lab

  2. Motivation Many tasks in computer vision and pattern recognition are based on local image features. Feature Extraction - Numerous feature extraction schemes have been proposed, like Harris Corner, SIFT etc. Similarity Measurement • However, similarity measurements for features are limited. Tianjin University Computer Vision Lab

  3. Motivation For similarity measurement (SM), two factors, i.e. & need to be considered. # correct matches/#total matches However, recently proposed SMs only improve the matching score but neglect the importance of the num of correct matches Tianjin University Computer Vision Lab

  4. Motivation The neglect inspires us to propose an effective similarity measurement for Tianjin University Computer Vision Lab

  5. Motivation There are 39 hits (matching score 85.97%))using the original matching method(OMM). There are 216 hits (matching score 93.11%)using our method(T-CM). Tianjin University Computer Vision Lab

  6. Triangle-Constraint Measurement Seed Point Selection – Bi-matching Illustration of bi-matching method. The matches (seed points) are those marked by ellipses. Tianjin University Computer Vision Lab

  7. Triangle-Constraint Measurement Organization of Seed Points Seed point False positive match Illustration of the Delaunay algorithm Tianjin University Computer Vision Lab

  8. Triangle-Constraint Measurement Triangle-Constraint PA Pi PB = * Illustration of the Triangle-Constraint Tianjin University Computer Vision Lab

  9. Triangle-Constraint Measurement Triangle-Constraint S S the descriptors for the Pi and the Cj respectively Radius of candidate area R A set containing all the temporary matches for PA and PB the Euclidean distance between the Cj and the Pi A predefined threshold Pe To handle the problem of false positive matches survived from Bi-matching, an additional step is taken after processing all the features from PA: The similarity score between the Pi and the candidate feature Cj is measured by If the maximum score of all the features in C is greater than a predefined threshold τ, the corresponding feature pair is considered as temporary match. To decide whether the temporary matches are accepted as final matches or not. C Illustration of the Triangle-Constraint Tianjin University Computer Vision Lab

  10. Experiment Evaluation Dataset – INRIA dataset Tianjin University Computer Vision Lab

  11. Experiment Evaluation Evaluation Criterion • The criterion of our evaluation is based on the number of • correct matches and the matching score. • - A match is defined as correct if the distance between the ac- • curate location and the estimated location is less than • 6 pixels, incorrect otherwise. Tianjin University Computer Vision Lab

  12. Experiment Evaluation Results – Relative image pair matching Tianjin University Computer Vision Lab

  13. Experiment Evaluation Results – Relative image pair matching Due to the huge amount of features that increasesthe possibility of accidentally considering incorrect matches as correct. Since the matching score is undefined (0/0) Tianjin University Computer Vision Lab

  14. Experiment Evaluation Results – Irrelative image pair matching There are 22 hits by the OMM and 0 hit by our method. Tianjin University Computer Vision Lab

  15. Conclusion • Triangle-Constraint Measurement: • Effective technique for similarity measurement to • improve both the number of correct matches and • the matching score. • Invariant to translation, rotation , scale and affine • transformations. • Robust to partial perspective distortions. Tianjin University Computer Vision Lab

  16. Question Questions? Computer Vision Lab @ TJU Tianjin University Computer Vision Lab 2009-08-11

  17. http://cs.tju.edu.cn/orgs/vision Thank you very much! Computer Vision Lab @ TJU Tianjin University Computer Vision Lab 2009-08-11

More Related