Handwritten numeral recognition by tree classification methods

摘要

The development of a binary tree for recognition of hand-written numerals using topological features is discussed. The algorithm uses a binary tree operating at two hierarchical levels. In the primary level, subclasses are derived using a set of ‘border-transition features’. In the second level, individual characters are recognised through detection of those properties unique to a given numeral. In an extensive test with ‘normally’ handwritten numerals, the algorithm yielded 98.8% accuracy.