利用方向性次頻分解之不受大小比例影響的紋理影像辨認

Abstract

這篇論文是說明如何利用方向性次頻分解做出不受大小比例影響的紋理影 像辨認.這種次頻分解是用一種濾波器將頻譜作方向性的切割,因此其輸出 可提供方向性的訊息.從經過次頻分解的紋理影像中,我們推導出兩種不受 比例影響的影像特徵,一為正規化功率,一為正規化相關性,在辨認時,將未 知的影像特徵取出,與事先建立好的資料庫比較,選取最接近的作為辨認結 果.經實驗證實,利用我們的方法,辨識率可達 98%. These thesis proposes a scale-invariant texture classification scheme by using directional subband decomposition. The decompo- sition is characterized by a bank of directional subband filters that allow a two-dimensional input signal to be represented by a sum of maximally decimated subband images and perfectly recon- structed from these decimated ones. In each decomposed channel image, we derive scale-invariant features which correspond to the normalized power and the normalized correlations. Training images are used to find feature templates. During classification , the unknown texture is matched against all the templates and the best match is taken as the classification result. From simu- lations, we find that the highest classification rate using 16 band decomposition for 16 kinds of texture is 98%.