| Summary: | 碩士 === 中正理工學院 === 電子工程研究所 === 82 ===
The aim of this thesis is to derive invariant features which are based on the homomorphic system and extracted from the spectrum histogram of the object image. Three kinds of feature models are proposed and each model is derived from the combiation of central moment and standard deviation. The features in each model are changed following a variable k to search for the best feature of an object. All these three models are named homomorphic relative moment features.
In order to check the performance among all of the features and compare with other types of features, this thesis brings up the stability and departure as the performance index. By testing on some adverse conditions, it can be found that the homomorphic relative moment features not only have good stability on scaling, rotation, environment illumination changing, image blurring, and noise corruption, but also have quite departure among different objects comparison with the well-known invariant moments. Therefore, the homomorphic relative moment features in this thesis may have good performance on tracking and pattern recognition. Also, the computing time for homomorphic relative moment features is less than that for the invariant moments.
|
|---|
