Formulation of invariants for discrete orthogonal moments and image classification / Pee Chih Yang
The aim of this thesis is to study invariant algorithms for the domain of discrete orthogonal moments. The invariants are anisotropic scale and translation invariants, translation, rotation and scale invariants, and affine moment invariants. Due to the complexity of hypergeometric functions, exis...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2013
|
| Subjects: | |
| Online Access: | http://studentsrepo.um.edu.my/4966/1/thesis%2Dpcy.pdf http://studentsrepo.um.edu.my/4966/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The aim of this thesis is to study invariant algorithms for the domain of discrete orthogonal
moments. The invariants are anisotropic scale and translation invariants, translation,
rotation and scale invariants, and affine moment invariants. Due to the complexity of
hypergeometric functions, existing invariant algorithms are slow. In addition, some of the
features have poor classification performance and are highly sensitive to the noise. Therefore
new sets of invariant algorithms have been proposed to resolve the above mentioned
issues. Discrete Tchebichef moments are selected as the implementation platform of the
proposed algorithms.To evaluate the performance of invariant algorithms, empirical studies
have been carried out on large set of binary images which consist of numbers, English
letters, symbols, Chinese characters and objects like animals, trees and company logos
under noiseless and noisy conditions. The discriminative power of the features is studied
using a set of very similar handwritten Chinese characters. Experimental results showed
that the proposed invariant algorithms are superior in computational efficiency. The generated
features from the proposed invariants in general, demonstrate improvements in
classification performance in both noiseless and noisy conditions |
|---|