利用形態梯度及遺傳演算法做階層式的紋理影像分類

Abstract

本論文提出一個以結合形態梯度及遺傳演算法為基礎的階層式紋理影像分 類法。這方法分成兩個階段：學習階段和分類階段。在學習階段，我們利 用遺傳演算法去找一組最佳的結構元素以提供最好的紋理分類，除此之外 ，我們並根據所選取出來的結構元素而建構一個決策樹。在分類階段，我 們利用走訪決策樹的方式來達成紋理影像分類的目的。實驗結果顯示，我 們所提出的方法只要使用很少而有效的結構元素便可達到很高的分類正確 率。 This thesis proposes a novel method of hierarchic textures classification using morphological gradients and genetic algorithms. The method consists of two phases: the learning phase and the classification phase. In the learning phase, genetic algorithms are employed to find an optimal set of structuring elements which provides the best discrimination of textures. The association of these structuring elements is then formed via a binary decision tree. During the classification phase, the textures are classified by simply traversing the constructed decision tree. The thsis also shows experimental results to demostrate that, by employing only a few effectual structuring elements, the hierarchic method achieves high classification accuracy and efficiency.