利用紋理合成及形態學運算做彩色影像修補之研究

Abstract

為了修補影像中受損的區域，或是移除影像中的某個物件，常常必須使用影像修補的技術。因此如何有效且正確的修補一塊未知的區域便成為一個重要的研究領域。本論文嘗試使用紋理合成及形態學運算來做影像修補。實作上先使用數學形態學的Opening將受損區域分為兩大部分，其中一部分是影像中受損極小的區域，另一部分則是受損範圍較大的區域。針對受損極小的區域，計算周圍可用灰階值的平均值進行修補。而針對受損範圍較大的區域，則尋找影像中最相近的紋理來修補，同時，應用FFT區塊比對法去加速搜尋的時間。實驗結果顯示我們所提出的方法不僅能獲得令人滿意的修補品質，且在運算的時間上是相當有效率的。

Inpainting, techniques can repair damaged photographs and remove/replace selected objects on the image. The goals and applications of inpainting are numerous. How to effectively and correctly inpaint an unknown region has become an important research topic. In this thesis, we propose a new inpainting method by using texture synthesis and morphological operations. Based on mathematical morphology, we first use opening to split damaged regions into two parts. One consists of small damaged regions and the other consists of large damaged regions on the image. We inpaint the pixel in the small damaged regions by computing the average of undamaged neighboring pixels surrounding it. For the large damaged regions, we search the most similar texture to synthesize it. Simultaneously, we apply the FFT block matching algorithm to speedup the time of searching similar textures. The proposed method is computational efficient. Experimental results look “reasonable” to the human eye.