基於臉部表情圖片之唱歌人像動畫系統

Abstract

本論文實作完成一套人臉歌唱動畫系統。此系統呈現出人物跟隨著背景音樂來唱歌，並且同時能有臉部表情的變化。我們使用多張臉部表情圖片作為表情資料庫，來合成出臉部的情緒表現。採用的表情圖為Paul Ekman博士所提出的六大基本情緒：快樂﹙happy﹚、憤怒﹙anger﹚、害怕﹙fear﹚、驚訝﹙surprise﹚、哀傷﹙sad﹚、厭惡﹙disgust﹚。歌唱時的發聲嘴形則藉由文字轉換語音系統，Microsoft Speech SDK，來找出相對應於歌詞的嘴形。接著利用文章中所敘述的方法來將表情圖與發聲嘴形做合成，即完成動畫中關鍵影格﹙key-frame﹚的製作。關鍵影格與關鍵影格之間的連續畫面，使用圖形漸形合成﹙image morphing﹚的技術來產生。將這些畫面串連起來，就是栩栩如生的人臉唱歌動畫。此系統除了能降低製造擬真歌唱動畫的成本，在娛樂性的應用方面也相當具有效果。

In this thesis, we implemented a singing-face animation system. In this system, the character sings a song with background music and can change his/her facial expressions simultaneously. We use several expression images to create facial emotional expressions. The images in emotional database are with six basic emotions: happy、anger、fear、surprise、sad and disgust, proposed by Dr. Paul Ekman. The visemes mapping to lyrics are obtained by a text-to-speech system, Microsoft speech SDK. Then, the approach proposed in this article is adopted to synthesize an expression image and a viseme. After that, generating a key-frame of the animation is finished. Using image morphing technique can generate the frames between two key-frames. These clips form a mimic singing-face animation. The proposed system not only lowers the cost of producing a realistic singing-face animation, but also quite powerful for recreational applications.