Visualizing the Distribution of a Large-Scale Pattern Set using Compressed Relative Neighborhood Graph
Masanori Goto, Ryosuke Ishida, Seiichi Uchida · IEEJ Transactions on Electronics Information and Systems · 2017
The goal of this research is to understand the true distribution of character patterns. Advances in computer technology for mass storage and digital processing have paved way to process a massive dataset for various pattern recognition problems. If we can represent and analyze the distribution of a large-scale pattern set directly and understand its relationships deeply, it should be helpful for improving classifier for pattern recognition. For this purpose, we use a visualization method to represent the distribution of patterns using a relative neighborhood graph (RNG), where each node corresponds to a single pattern. Specifically, we visualize the pattern distribution using a compressed representation of RNG (Clustered-RNG). Clustered-RNG can visualize inter-class relationships (e.g. neighboring relationships and overlaps of pattern distribution among “multiple classes”) and it represents the distribution of the patterns without any assumption, approximation or loss. Through large-scale printed and handwritten digit pattern experiments, we show the properties and validity of the visualization using Clustered-RNG.