經由照片資料庫實現繪圖與卡通人像真實化之研究

Abstract

要將不真實的卡通人像或是畫像真實化是一件很困難的事，對於人臉結構方面需要高度的想像力以及理解能力。人眼的高敏銳度更使得人臉真實化更為困難。根據我們的調查，目前對於從卡通人像半自動或是自動化去產生出真實化的樣貌的研究非常的少。如果我們使用目前有的方法直接warping或是matching，會使很多較為精細的資訊找不到且亦可能會使重要的個人特徵遺失。在這篇論文當中，我們提出了一個example-based方法來合成真人化的卡通人像或是畫像。我們使用graph-cut-based optimization來拼出和輸入圖最相似且合適的piece-up臉。下一步，我們更提出了chromatic gain compensation和multi-level blending的方法來無縫的拼貼patches。我們的實驗結果更顯示我們提出的example-based method是能夠對於輸入的卡通圖產生出真實且新穎的結果。

Realizing unrealistic cartoon faces or paintings is a complicate task required high imagination and comprehension of face structures. The high sensitivity of human eyes makes face realizing a difficult problem. According to our survey, there are rare works on fully- or semi-automatic generation of a realistic face from a cartoon face. When directly applying face warping or matching methods, existing methods does not have much detailed information or may lose the personally characteristics. In this thesis, we propose an example-based method to synthesize the reality of cartoon faces and portrait paintings. We use graph-cut-based optimization to find the most similar and adequate piece-up faces according to the input image. Next, we further present chromatic gain compensation and multi-level blending to seamlessly stitch the patches. Our experiments show that the proposed example-based method is able to provide realistic and novel result of input faces.