1. REFA: A Robust E-HOG for Feature Analysis for Local Description of Interest Points
- Author
-
Michel Dhome, Christophe Tilmant, Manuel Grand-Brochier, Institut Pascal (IP), and Université Blaise Pascal - Clermont-Ferrand 2 (UBP)-SIGMA Clermont (SIGMA Clermont)-Centre National de la Recherche Scientifique (CNRS)
- Subjects
Matching (statistics) ,Computer science ,business.industry ,Detector ,Scale-invariant feature transform ,Pattern recognition ,02 engineering and technology ,030218 nuclear medicine & medical imaging ,Image (mathematics) ,03 medical and health sciences ,0302 clinical medicine ,Histogram of oriented gradients ,[INFO.INFO-TS]Computer Science [cs]/Signal and Image Processing ,0202 electrical engineering, electronic engineering, information engineering ,020201 artificial intelligence & image processing ,Artificial intelligence ,business ,ComputingMilieux_MISCELLANEOUS - Abstract
This article proposes a Robust E-hog for Feature Analysis (REFA) to describe interest points and their neighborhood. Initially the two most used methods: SIFT and SURF are studied and various advantages (invariances, repeatability) are extracted to create a new approach (detection, description and matching). First, the Fast-Hessian detector is used because it gives the best repeatability rate, however it will be optimized. Secondly the local neighborhood description is based on a histogram of oriented gradients on an elliptical shape. Finally a decision tree, validation threshold and deletion duplicates are used to match interest points. This method must also be as robust as possible for image transformations (rotations, scales, viewpoints for example). All tool parameters (orientations, thresholds, analysis shape) will be also detailed in this article.
- Published
- 2013