A Method To Improve Interest Point Detection And Its Gpu Implementation
Interest point detection is an important low-level image processing technique with a wide range of applications. The point detectors have to be robust under affine, scale and photometric changes. There are many scale and affine invariant point detectors but they are not robust to high illumination c...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | en |
| Published: |
LSU
2012
|
| Subjects: | |
| Online Access: | http://etd.lsu.edu/docs/available/etd-11132012-154312/ |
| Summary: | Interest point detection is an important low-level image processing technique with a wide range of applications. The point detectors have to be robust under affine, scale and photometric changes. There are many scale and affine invariant point detectors but they are not robust to high illumination changes. Many affine invariant interest point detectors and region descriptors, work on the points detected using scale invariant operators. Since the performance of those detectors depends on the performance of the scale invariant detectors, it is important that the scale invariant initial stage detectors should have good robustness. It is therefore important to design a detector that is very robust to illumination because illumination changes are the most common. In this research the illumination problem has been taken as the main focus and have developed a scale invariant detector that has good robustness to illumination changes.
In the paper [6] it has been proved that by using contrast stretching technique the performance of the Harris operator improved considerably for illumination variations. In this research the same contrast stretching function has been incorporated into two different scale invariant operators to make them illumination invariant. The performances of the algorithms are compared with the Harris-Laplace and Hessian-Laplace algorithms [15].
|
|---|