| Paper: | IMDSP-L1.2 | ||
| Session: | Image and Video Analysis | ||
| Time: | Wednesday, May 19, 13:20 - 13:40 | ||
| Presentation: | Lecture | ||
| Topic: | Image and Multidimensional Signal Processing: Image and Video Analysis | ||
| Title: | A NEW AND COMPACT ALGORITHM FOR SIMULTANEOUSLY MATCHING AND ESTIMATION | ||
| Authors: | Hongdong Li; Australian National University | ||
| Richard Hartley; Australian National University | |||
| Abstract: | Feature matching and transformation estimation are two fundamental tasks in video processing and computer vision research. They are also two interlocked problems: one’s solution is often the other’s precondition. This paper shows that they are in deed two mathematical dual problems. By applying such duality, a compact algorithm of less than 10 lines pseudo-matlab codes, which is able to solve the two interlocked problems simultaneously, is provided. We show that the solutions of matching and estimation are simply two matrix factors of two Grammians. A Newton-Schulz numerical algorithm is applied to solve a factorization problem. Both correspondence and transformation are updated in an alternative way. Both starting from random initial guesses, and soon converge to correct solutions. The effectiveness of the proposed algorithm is illustrated from experiments on both synthetic and real images. This algorithm can also be extended to high-dimensional feature sets without any modification. | ||
| Back | |||
Home -||-
Organizing Committee -||-
Technical Committee -||-
Technical Program -||-
Plenaries
Paper Submission -||-
Special Sessions -||-
ITT -||-
Paper Review -||-
Exhibits -||-
Tutorials
Information -||-
Registration -||-
Travel Insurance -||-
Housing -||-
Workshops