摘要
影像修補技術主要修復影像中不清晰或受損的區域，或是移除影像中的某個物件，因此如何有效且正確的修補一塊未知的區域便成為重要的研究議題。由於人眼判定物體或影像中的內容自然與否，主要是由輪廓來做判斷，所以本研究先對影像的結構進行修補，將影像的結構勾畫出來，再進行材質的修補，以達到更好的修補品質。本研究的整合方法(Mashup Method, MM)提出先修補結構再修補材質的修補架構(framework)，主要的概念先修補架構，接著使用邊緣偵測保留修補完成的結構，再利用MM方法產生一張新遮罩，最後對新遮罩的區域做材質修補。本研究的實驗結果顯示，MM方法的PSNR值與自然度都維持在最高，而且修補時間成長幅度不大，說明MM方法的修補品質在客觀數據與人眼主觀感受的條件下都有好的表現。
並列摘要
In order to inpaint images which are not clear and with some damaged areas, or to remove the image of an object, image inpainting techniques are often adopted as a major technique. Therefore, how to inpaint an unidentified zone effectively and correctly has become a crucial research topic. For the reason that recognizing the content objects or images by human eyes is through the contour of the objects, hence the inpainting process might be effective if we begin with inpainting the structure of the image, and next sketch the image of the structure.The research provides a method of image inpainting-Mashup Method. Firstly we use the inpainting methods which is favorable for image structure to do the first inpainting; secondly, we use edge detection (Sobel filter) to retain the good structural information, and then implements the merge method that is proposed in this paper to produce a new image; finally the image texture which enables inpainting method would be implemented to inpaint some part of image texture as the last step. Compared with the previous methods, the proposed method in the research reveals a good performance in both running time and the quality of the images.