Educational Character Recognition System Implementing an Interactive Visualization of Multi-Dimensional Distribution

我们提出了将多维分布可视化应用于模式识别系统，并构建了一个多维空间的教育可视化系统，刺激了人们对模式分布的直观和可视化的理解。研究了手写体数字识别系统中每个字符多维向量的可视化方法，通过主成分分析将原始向量转化为主成分。每个类别的概率相等的椭球体被导出，并在可视化系统中显示的三维空间中可视化。结果，尽管可视化的分布并不代表所有的向量，但是在简化的三维空间中可以容易地看到每个类别分布的边界表面。总之，使用该系统可以激发模式识别学习者对多维分布的兴趣和理解。

We propose visualizing multi-dimensional distribution used in pattern recognition system and construct an educational visualization system for multi-dimensional space which stimulates intuitive and visual understanding of pattern distribution. We treat the method of visualizing the multi-dimensional vectors of each character on the handwritten digit recognition system.By applying principal component analysis, the original vectors are converted into the principal components. Ellipsoids of equal probability for each category are derived and visualized in the three-dimensional space shown in the visualization system. As a result, the boundary surfaces of each distribution of categories can be seen easily in the reduced three-dimensional space in spite of the fact that the visualized distributions do not represent all vectors. In conclusion, learners of pattern recognition can be stimulated their interest and comprehension of multi-dimensional distribution using this system.