中文摘要：

为三维模型创建纹理贴图是真实感图形学领域的一个重要课题,然而现今针对大数据的复杂结构的自然场景的纹理映射方法还很少.该文提出了一种新的多视图纹理映射方法,能够利用多角度拍摄的图像序列为三维模型进行高真实感的纹理贴图.相比现有的方法,该文的方法在针对模型中存在复杂遮挡的纹理映射以及大纹理数据的融合这两方面有着明显的优势.首先将多视图纹理映射转化为一个马尔可夫随机场问题,并用能量优化的方法获得模型每个面片最优的纹理图像对应关系;然后将映射的纹理数据进行精简化,并对部分不可见区域的无效纹理进行修复;最后,利用梯度融合的方法消除纹理拼接边界的颜色差异,最终获得无缝隙的纹理贴图效果.一系列复杂场景的实例证明了该文方法的有效性和鲁棒性.

英文摘要：

Texture mapping of 3D models is an important problem in photorealistic graphics.However,few methods are proposed to deal with large-scale texture mapping of natural scenes with complicated structures.This paper presents a novel multi-view texture mapping method,which can recover photorealistic texture maps for 3D models using multi-view images.Compared to the existing methods,our method has obvious advantages in texture mapping with complicated occlusions and fusion of large-scale texture data.We first formulate multi-view texture mapping as a Markov Random Field problem,and minimize the energy function to get an optimal texture image map for each face on the 3D model.Then,we simplify the mapped texture data,and restore the invalid textures in some invisible regions.Finally,we eliminate color differences along the texture stitching boundaries by gradient fusion to achieve seamless texture mapping.A variety of challenging examples demonstrate the effectiveness and robustness of the proposed method.