基於影像集之繪畫場景重現

Abstract

在照相機等攝影記錄工具還沒有被發明的時代，繪畫一直作為紀錄人物或是美麗風景的主要方式，同時也用來敘述許多重要事蹟。通常這些繪畫都是由專業的藝術家所繪製，因此在顏色與結構等的表現上也相當重視，作出來的作品也大多都符合了人類的美感。 在現代，我們則是利用照相機或攝影機等來拍攝記錄我們周遭的生活。但是對於那些藝術家所畫的繪畫，如果我們想要照出一模一樣的照片基本上是不太可能的。直接利用一般的繪圖軟體去轉換也有一定的難度，無法輕易的將繪畫調成具有真實度的'照片'。 因此在這篇論文中，我們提供了一個方法讓繪畫能夠轉換成類似照片的結果。藉由使用者所提供的與繪畫有部分區塊相似的照片，以及相對應的提示筆畫，根據這些對照片作具結構感知的形變。最後再利用繪畫所含的資訊將這些圖片合成為擁有照片感覺的結果。

Before the widespread of modern capture devices, painting played an important role in depicting and recording the real world for human. These paintings provide immediate depictions of the real world, and usually have artistic appearances. While modern devices aim at recording the world in the form of photorealistic appearance, it is challenging to produce a photorealistic counterpart of a painting. In this paper, we present a method for ``transferring' an input painting into a photorealistic appearance by referring several photos which are partly similar to the painting. With a number of user-specified corresponding strokes on both the painting and photos, the ``transferred' result is generated by deforming these corresponding strokes with a structure-aware warping technique and composing the warped photo regions with the painting-guided compositing technique. We demonstrate the effectiveness of the proposed method on a variety of paintings.