本篇論文提出一個結合色彩、紋理及區塊資訊來分類影像的方法。首先,利用色彩及紋理特徵將影像作前置性的分割處理,並透過區塊的區域資訊,確切地表達出影像中所包含的物件。在檢索的過程中,先利用查詢影像的整體資訊初步篩選出和查詢影像較為相似的資料庫影像,然後再針對所篩選出的資料庫影像比較其和查詢影像間區塊所具有的色彩紋理特徵的相似程度,並將篩選出的資料庫影像加以排序,最後再利用k-NN的規則達到影像分類之目的。由實驗結果證實本法確實可行。
A new classification method by integrating color, texture and region is proposed in this study. We adopt a color texture segmentation method that unifies color and texture features to obtain semantic regions. Image-based features related to color and local edges patterns are then used to prune irrelevant database images for each query image. The proposed region matching is then applied to each pair of the query image and the database images in the small plausible set. Thus, the dissimilarity of each pair can be calculated on the basis of the matching results. Finally, all the database images in the plausible set can be ranked in the ascending order of dissimilarity values. To achieve the classification goal, the k-NN rule is used to assign class label to the query image. The effectiveness and practicability of the proposed method has been demonstrated by various experiments.