Efficient Pre-Computation for Bispectrum Based Features for Rotation and Translation Invariant Image Recognition
In this thesis we explain the use of an approximate but fast algorithm for the computation of discrete spherical Fourier coefficients. These coefficients have been used for the computation of a set of translation and rotation invariant features used to classify handwritten digits. We have trained su...
Saved in:
| Main Author: | |
|---|---|
| Format: | Dissertation |
| Language: | English |
| Published: |
ProQuest Dissertations & Theses
01-01-2011
|
| Subjects: | |
| Online Access: | Get full text |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In this thesis we explain the use of an approximate but fast algorithm for the computation of discrete spherical Fourier coefficients. These coefficients have been used for the computation of a set of translation and rotation invariant features used to classify handwritten digits. We have trained support vector machine classifiers using the directly computed features and those computed using the approximate coefficients and compare classification errors. We have examined the relative errors between the directly computed coefficients and the approximate coefficients, the relative errors between features, and the relative computational times for each set of coefficients. We found that the features computed using the approximate coefficients achieve approximately the same classification errors with less computations. Further, the time savings of using the approximate algorithm grows as the sizes of the images increase. |
|---|---|
| ISBN: | 1267174579 9781267174574 |