Collection of on-line handwritten Japanese character pattern databases and their analyses

We describe the design of on-line handwritten Japanese character pattern databases, software tools for pattern collection and verification, and analyses of collected patterns. Two databases containing over 3 million patterns were compiled: one with 120 people contributing 12,000 patterns each and another with 163 participants contributing 10,000 patterns each. Patterns were collected mostly in their sentential contexts and verified by machine and human inspection. Their analyses reveal greater variations in stroke count for characters having many strokes, with people generally using fewer strokes; they additionally reveal that stroke order variations are generally caused by common habits and added strokes.