【作者】 徐源；

【导师】 徐庆；

【作者基本信息】 天津大学 ， 计算机应用技术， 2006， 硕士

【摘要】 在真实感图像绘制的研究中有两个主要的发展方向。其一是提高绘制的真实感,真实地再现各种复杂的光照场景。另一个就是在图像质量与计算量之间取得平衡,用尽可能少的计算量得到尽可能高的图像质量。我们的研究就聚焦于第二种情况。自适应技术就是为了应对这种要求而提出来的技术。自适应技术通过将采样数目不平均地分配到各个像素上来达到目的。即从初次采样后得到的图像中挑出质量不好的区域,再追加计算量。自适应技术已经成为满足这种要求的主要技术手段。而在这种方法中,评价质量好坏的标准是最关键的部分。除了自适应技术,progressive rendering也是一种需要平衡图像质量与计算量的技术。在这篇论文中,我们对自适应、质量评测标准给出了详细地介绍。同时还对一种经典progressive rendering算法进行了详细介绍。从这三方面的知识中,我们提出了自己的组合测量标准,并把它应用到自适应方法中。同时我们改变了自适应采样中最常用的控制方法,代之以四分原始图像平面的技术。我们通过大量的实验较为全面地探讨了这种新方法的表现。实验的结果图像与超采样所得的结果求RMS值。通过这个值与对应的平均采样点数就能看出算法的优劣。通过这样的方法我们将实验结果与对应的经典对比度控制的自适应技术进行可比较。我们对最后的结果中的优点和不足之处给出了详细地讨论,并对将来的发展提出了建议。更多还原

【Abstract】 There are two goals for realistic image synthesis technology. One is to produce pictures that exactly like a photo produced by a cammare; It can simulate all kinds of illuminating conditions and guarantee a realistic result. The other is to gain a balance between the image quality and computation price. We want good images as less computation as possible. We dedicate to the second goal. One preferable chosen is the adaptive sampling technology. The adaptive sampling technology can distribute computation according to the quality of different areas in image plane. After a low sample rate, we get a low quality image. The method can find where the quality is not enough, and add samples to the area. The key to the adaptive sampling technology is a good measurement of pixel quality.Besides the adaptive sampling technology, progressive rendering also needs a balance between the image quality and computation price. We introduce the konwdege about adaptive sampling, quality measurement, and a classic progressive rendering technology. We get inspiration from the konwdege and prompt our own quality measurement, and practice it in adaptive sampling technology frame. We also change the control flow and replace it with a quadtree from the original image plane. We study all spects of the performance of the new algorithm by experiments. We compare the RMS of the result image and the standard image produced by heavily sampling. The RMS and the corresponding average sampling number denote the performance of the algorithm. We compare the algorithm with the classic contracted based adaptive sampling in the way. We study the advantages and the shortages. At last we propose advices for the development of the algorithm.更多还原