以影像為基礎之立體人頭合成

Abstract

利用二維影像來合成三維物件的研究一直是很受到重視的一個領域，在本論文中，我們提出一個三維立體人頭的合成方式，以多張不同視角的影像資訊為基礎，配合一個事先準備好的通用三維人頭模型(generic 3D head model)，透過二維影像中特徵點的選取來估算出拍攝的角度、方位，進而處理人臉模型五官的資訊重塑與特徵點對應的關係，接著利用內差技術，使用放射基底函數(radial basis function)對三維通用人頭模型作形變(morphing)，使形變過後人頭模型的形狀與真實人臉相符，最後對三維通用人頭模型作材質貼圖(texture mapping)，並可對此模型作任意角度、方向的旋轉平移，以期達到類似照鏡的功能。

The research of making use of 2D image to form the 3D objects has been a field paid attention to very much all the time. In this work, we propose the synthetic way of a 3D human head, based on image information of multiple-view, cooperate with a Generic 3D head model. By choosing the feature points from images to estimation of position of the camera, and then dealing with information reconstruction of human head and feature point relationship between images and 3D object. Subsequently, applying interpolation technique, the head model is deformed from original model to fit the computed feature points by using Radial Basis Function. Make the form of the head model of people after deformation in conformity with true face of people. Finally, execute 3D head texture mapping and reach the function of translation and rotation with arbitrary direction.