本論文提出一套具偵測並修復毀損邊緣之影像修補系統。首先找出可能的邊緣線端點,再根據顏色與切線斜率資訊,執行邊緣點配對。進而進行影像修補工作,包含毀損邊緣軌跡外圍區域與其它毀損區域之修補。由於不同影像具其個別特性,一個修補法無法針對所有影像做完美修補。因而我們根據不同的影像特性,針對邊線區域與非邊線區域提出四個不同修補方法。 修補法之一,以邊線為基準之條狀剪貼,即對兩邊邊緣點利用附近相關的非毀損區域資訊進行差異比對,將最小差異的條狀區段依序的貼至受損區域完成修補。 修補法之二,以邊線為基準之區塊剪貼,以兩端點邊線的延伸,找出適當的區塊,利用座標的平移與旋轉,將對應的座標點之內容資料填回受損區域完成修補。以上兩個修補法都執行相同的後處裡工作,即對修補區域邊緣的顏色差異,以區域變化平滑化(local variation smoothing)將之消除。 修補法之三,比較顏色差異之修補法,在受損區域由外至內一層一層的修補。對每一個待修補點,與其鄰近未受損點,比較其變異數,將變異數最接近之未受損點,將其相關資訊填入該受損點完成修補。 修補法之四,方向與顏色資訊之內插法,即對每一對邊緣點利用B-spline恢復並描繪出受損邊緣的軌跡。待所有的受損邊緣軌跡描繪出後,再以線性內插法來修補受損邊緣軌跡的外圍區域。對其它受損區域兩階段之修補法,則從最靠近非受損區域的待修補點開始,根據其鄰近非受損區域的特徵決定其修補的顏色。 對以上每個修補法,我們分析其特性並描述其所適合修補的影像,最後以實驗結果驗證之。 This paper proposes an image repairing system that can detect and restored damaged edges. First possible damaged edge endpoints are determined and then paired in accordance with their characteristics such as color information and slopes of tangent lines. The damaged region to be repaired is divided into two parts including the subregion around the predicted edge trajectory and the remaining. There does not exist an image repairing system well suit to all kinds of images. Based on different characteristics of edge regions and non-edge regions, we proposed four different repairing methods, including strip-wise cut-and-paste with adjustment, block-wise cut-and-paste with adjustment, inpainting based on color variation, and inpainting using interpolation on color information and direction. For each of these methods, we analyze and describe its individual characteristics and the images it is suitable to repair.