數位影像修補及修改技術是一種影像竄改的機制，這種技術可以自動地修補影片中被損毀的區域或是移除影像中的物件。大多數修補技術都使用單一解析度的方法，去推斷被破壞的影像像素資訊，然後進行修補。在本論文中，我們將提出一個使用多重解析度為基礎的演算法，針對不同階層的解析度進行考慮，提供一個影像修補的機制。這個技術的原理是將需要修補的影像切分為許多小區塊，並利用每一區塊與階層中色彩變異度的關係進行評估，決定可用來修補的資訊。 另外，特別針對傳統中國繪畫的特性，提出一個使用多重圖層為考量的修補策略，成功地結合多重解析度的影像修補演算法，針對不同圖層中的解析度進行考慮，並設計一個多重圖層的修補與圖層合併演算法。在此研究中，分別實驗1500張各類的圖片，包括卡通、風景照片、國畫與西洋畫等影像圖片，進行不同比例破壞程度之修補成效測試，並與不同的影像修補技術，進行修補後結果與效率之比較。根據實驗的結果，我們所提出的修補策略，其修補後的圖片，具有相當高的PSNR數值，其執行修補程序的效率，較其他方法快速。 Digital inpainting is an image interpolation mechanism, which can automatically restore a damaged or partially removed image. Since most inpainting mechanisms use a singular resolution approach on the extrapolation or interpolation of pixels, this thesis proposes a multi-resolution algorithm, which takes into consideration the different levels of detail. The algorithm is based on the concept of image subdivision and estimation of color variations. Noises inside blocks of different sizes are inpainted with different levels of surrounding information. The results showed that an almost unrecognizable image can be recovered with visually good results. Furthermore, we study how Chinese paintings are drawn, and propose a multilayer inpainting mechanism which can be used effectively on Chinese and western paintings. The thesis conducts a new approach, which divides a Chinese painting into several layers and each layer is inpainting separately. A layer fusion mechanism then finds the optimal inpaint among layers, which are restored one-by-one. We apply the algorithm on Chinese and western drawings. These algorithms were tested on 1500 still images and an evaluation shows the effectiveness of our approach, a high PSNR value as well as a high level of user satisfaction.
