彩色紋理影像切割

Abstract

本論文提出了一個新的彩色紋理影像切割法。這個方法包含了兩個主要的 部份：紋理特徵的選取，以及紋理影像的切割。首先，在紋理特徵的選取 方面，我們發現計算本論文所提出的紋理特徵，所花費的時間不多，而且 所選取的特徵亦頗具代表性。在紋理影像切割方面，我們則利用分群的技 巧來達到影像切割之目的。這個分群法是將高度空間的解析式二群分群公 式以先分割再合併的形式推導而得。實驗結果顯示，本文所提之法在切割 紋理影像時，不但比其它的方法快，而且所得到的分割效果也較佳。更重 要的是，我們不必事先知道一張影像含有幾種不同的紋理，因為方法本身 便能對此作一判斷。最後，我們將上述所選取的紋理特徵應用到紋理影像 分類及紋理邊緣的偵測。實驗結果顯示，在這些應用上，也都能得到良好 的效果。 In this thesis, we propose a new method to segment color texture image. This method consists of two main parts, feature selection and image segmentation. For feature selection, The features suggested in this thesis are found to have good descriptions about the characteristics of the given texture images. It is observed that the time needed for evaluating the selected features is shorter than that of the traditional features. As for image segmentation, we use a new clustering technique in this thesis to accomplish segmentation. This new clustering technique is a "split-and-merge" technique using some high diminsional two-class analytical formulas. An important advantage of the proposed segmentation technique is that the technique does not need to know the number of segmented regions contained in a image. This property is very useful for automatic image segmentation. The experimental results show that the proposed method can segment well the given color texture images, and the time needed is short. Finally, we also use the proposed features to detect texture edges or classify texture images. The experimental results show that the proposed method also yields good results and the computation time is short, too.