使用紋理特徵及區域成長法而得的影像切割技術

Abstract

在本論文中，我們提出了一個將影像的平滑區和紋理區分開處理的影像切 割方法。在一開始我們將影像經由 Sobel 方法及細線化方法處理得到一 張邊界線影像。然後我們使用區塊切割法將原始影像切割成一些不重疊的 區塊，然後將這些區塊分成兩群以得到兩張平滑區及紋裡區影像。有關平 滑區的影像處理，我們利用區域成長法，結合邊線偵測及邊線追蹤法，來 獲得平滑區的影像切割。有關紋理區影像的處理，我們先計算每一點的紋 理特徵值，再求每一點的「區域」紋理特徵值。我們使用區域紋理特徵值 來做紋理區影像切割。因為我們使用區塊切割法可將影像大致切割出來然 後再仔細處理邊緣區，所以我們不必事先知道要切割的群數。由於結合了 不同的技巧，所以可以獲得較佳的切割結果。 In this thesis, we propose an image segmentation method to process separatly the smooth regions and texture regions of an image. At the beginning, we preprocess the image with the Sobel operator and a thinning algorithms to obtain an edge image which can provide the boundary needed in region-growing. We then use the block partition method (BPM) to partition an image into non-overlapping blocks, and then classify the blocks into smooth or non-smooth blocks. First, we process the smooth blocks by combining region growing method with the boundary detection and tracking technique to obtain the segmentation result of the smooth regions. Then we process the texture part of the image by statistcal method to get the local texture features of each pixel. According to the feature values of each texture pixel, we use region-growing to obtain the segmentation image of the texture part. The advantage of our segmentation technique is that there is no need to know the cluster number, because we can use the BPM to obtain the rough segmentation result.