多層次的影像補繪

A Multi-Scale Approach For Image Inpainting

陳葦濃(Wei-Nong Chen)

指導教授 ： 鄭嘉慶

大同大學/電機資訊學院/電機工程學系所/碩士(2004年)

摘要

近年來影像補繪是一個相當熱門的研究課題。影像補繪主要是將影像中消失的資料，加以填補的動作。此研究可應用在許多領域，譬如照片或影帶的修補以及外加文字的去除等。
在這篇論文中，我們提出一個多層次的影像補繪方法。我們利用多層次演算法則之由粗略到精細的策略對補繪區域進行兩階段的補繪。在第一階段，首先，我們先將影像中已知區域分割為多個具有同質性的區域；接著利用與補繪區域相鄰的分割區域的紋理性質以及邊緣結構，對補繪區域內的曲線進行估測；藉此，我們可以找出已知區域中與補繪區域相鄰的分割區塊，它們消失在補繪區域內的區域範圍。關於第二階段的補繪，我們是對補繪區域內的每一塊分割區域，利用和它相鄰的外部分割區域內的已知資料進行補繪。我們採用一個以範例為基礎的方法，此方法藉著材質合成來補繪影像。以範例為基礎的材質合成法是從與目標區域相鄰的區域中找出所謂的範例影像資料，接著利用此範例影像資料來產生新的影像資料，然後用新產生的影像資料對目標區域進行填補；此外，填補區域的前緣上的點所對應的等輻透線向量值和區塊密度，將用來決定補繪的順序以達到填補擴張的目的。
此方法的主要目標是希望能夠精確地修復大型的線性以及非線性結構。我們成功的用一些真實的以及人工繪製的圖像來展現我們所提的方法，並獲得相當優異的實驗結果。

關鍵字

輪廓估測 ； 以圖素為主的補繪 ； 以區域為主的補繪 ； 影像補繪 ； 等輻透線

並列摘要

Image inpainting is an active research topic in recent years. The task of image inpainting is to fill-in the lose data in an image. It has many practical applications including the restoration of old photographs and damaged films, and the removal of superimposed text, etc.
In this work, we present a multi-scale image inpainting method which utilizes the multi-scale coarse-to-fine strategy to develop a two-stage inpainting technique. In stage one, we first segment the known region of image into several homogeneous regions, which results in a few contours with one endpoints connected to the border of blank region yet to inpaint. Then we utilized that the attributes of these contours including texture and contour characteristics to estimate the contour formation in the blank region. This results in a few segmented sub-regions inside the blank region. In stage two, we inpaint each of the sub-region resulted from the previous stage. We adopt an exemplar-based technique which inpaints an image via texture synthesis. The exemplar-based texture synthesis borrows example image data from the proximity of the destination region to generate new image data for filling in the destination region. In addition, the order in which the filling propagates is subject to filling priority which depends on the isophote and patch density of a point on filling front.
This approach aims at recovering large linear and non-linear structures. We successfully demonstrate our algorithm with several realistic and artificial images, and obtain promising experimental results.

並列關鍵字

,coutour estimation ； Image Inpainting ； Isophote ； Pixel-based inpainting ； Region-based Inpainting

參考文獻

[1] A. Hirani and T. Totsuka, “Combining frequency and spatial domain information for fast interactive image noise removal,” SIGGRAPH on Computer Graphics, p.p. 269-276, Aug. 1996.

[3] C. Ballester, M. Bertalmio, V. Caselles, G. Sapiro, and J. Verdera, “Filling-in by joint interpolation of vector fields and gray levels,” IEEE Transactions on Image Processing, vol. 10, no. 8, p.p. 1200-1211, Aug. 2001.

[6] A.C. Kokaram, R.D. Morris, W.J. Fitzgerald, and P.J.W. Rayner, “Detection of missing data in image sequences,” IEEE Transactions on Image Processing, vol. 4, p.p.1496-1508, Nov. 1995.

[7] A.C. Kokaram, R.D. Morris, W.J. Fitzgerald, and P.J.W. Rayner, “Interpolation of missing data in image sequences,” IEEE Transactions on Image Processing, vol. 4, p.p.1509-1519, Nov. 1995.

[9] Y. Deng and B.S. Manjunath, “Content-based search of video using color, texture, and motion,” in Proceedings of International Conference on Image Processing, vol. 2, p.p. 534 –537, 1997.