Sequential and mass entry type ancient document input system based on similar pattern retrieval

摘要

We propose an efficient ancient document input system based on pattern recognition. Heretofore, character recognition is based on the following two steps: (1) automatically cutting out a character pattern from an image; and (2) recognizing the character using a dictionary (template) created by learning. However, in an ancient document, the cutout of one character pattern is difficult due to contact between characters, and template creation based on the learning is substantially impossible because of the large number of character types. Accordingly, a conventional method cannot deal with an ancient document. In order to resolve these problems, we introduce a method that repeats the following two steps: cutting out a character pattern from an image by an operator; and mass entry of the same characters obtained by retrieving over the entire document using the character as a template. Experimental result showed the system could reduce the number of operations by 40% compared to the method an operator input one character at a time.