Efficient Completion for Multi-Dimensional Real-World Data

三维空间数据、图像、视频等多维数据正在快速增长。本研究旨在通过分析多维数据和提取特征值，开发一种新的自然完备化方法，而不是简单的插值法。效率是实现自然完备化的最重要的方面之一，因为分析海量的多维数据至关重要。该项目尝试了两种方法来实现这种效率，即“使用SIMD处理器进行高效计算”和“通过统计算法进行高效计算”。对于前者的尝试，我们将重点放在了海量数据的渲染上，并成功地完成了高效的实现，使GPU的性能得到充分发挥。对于后一种尝试，我们进行了面部印模手术的实验。我们成功地实施了保留原始脸部外观年龄的面部美容方法。

Multi-dimensional data such as 3D spatial data, image, video and so on is rapidly increasing. This research aims at developing a novel method of natural and completion instead of simple interpolation by analyzing multi-dimensional data and extracting characteristic values. Efficiency is one of the most important aspect to achieve natural completion because it is crucial to analyze huge multi-dimensional data. The project tried two approaches to this efficiency, namely, "efficient computation with SIMD processor" and "efficient computation by statistical algorithm". As for the former attempt, we focused on rendering of massive volume data and succeeded to accomplish the efficient implementation which allows GPU's full performance. As for the latter attempt, we experimented on facial-impression operation. We have succeeded to implement facial beautification method which preserves apparent age of the original face.