卡通化3D人臉產生系統

建立卡通化3D人臉在電腦圖學領域中是一個有趣和實用的研究題目。然而，仍舊缺乏在保持重要特徵的條件下，將2D的人臉影像轉成3D卡通人臉的工具大部分被提出來做卡通人臉和2D人臉影像合成的方法。在前置處理步驟，臉部特徵和輪廓通常使用影像處理的方法來擷取，例如使用邊緣檢測的方式來處理。接著，將人臉部分轉換成具有特定風格的NPR(Non-Photorealistic Rendering)圖像，例如:將人像照片轉換成水彩畫、油畫或鉛筆素描等等具有不同筆觸效果的肖像畫。在這篇論文中，我們提出一個不同的方法來將2D的臉部影像轉換成3D的卡通臉部模型，這個3D的卡通模型對於製造和客製化3D Q版可愛公仔將會非常的實用。我們使用DXUT (The DirectX Utility Library) 架構來開發使用者介面，而且採用了Open CV( Open Source Computer Vision Library ) 這一套電腦視覺函式庫來自動偵測臉部特徵的位置。然後，讓使用者藉由點選和移動特徵點來調整臉部的網格模型。當把臉部網格模型調整到符合所輸入的臉部影像，原始基本模型就會被調整和變形為相對應的臉部模型。眼睛和嘴巴的外型也被提取出來，並使用紋理映射來合成類似卡通的影像。

關鍵字

人臉卡通化；人臉特徵點擷取

並列摘要

Modeling 3D caricaturing face is an interesting and practical research topic in the field of computer graphics. However, there is still lack of tool to transform a 2D human face image into a 3D comic face with the important features been preserved. Most of the methods proposed for caricaturing face synthesize 2D facial images. In the pre-processing steps, facial feature and facial contour are extracted by adopting image processing methods such as contour detection. The input image is then transformed into the caricature image with a specific NPR (Non-Photorealistic Rendering) style, such as watercolor, oil painting, or pencil sketch. In this paper, we propose a different approach which converts a 2D face image into a 3D caricaturing model. This 3D caricaturing model can be useful for making and customizing 3D cute figures. We use the DXUT (The DirectX Utility Library) framework to develop a user interface, and adopt the computer vision library - Open CV ( Open Source Computer Vision Library ) to automatically detect important positions of several facial features. And then, users are asked to click and drag the feature point to adjust a controlling facial mask. After the controlling facial mask has fit the input facial image, the corresponding mesh model is determined and deformed. The shape of eyes and mouth is also extracted to synthesize cartoon-like images which are used later for the purpose of texture mapping.