For recognition of hallmarked characters on metal surfaces of various equipments, shadow image of characters by appropriate lighting is used. In actual environments where lighting conditions are limited, indistinct shadow images with noises make recognition difficult because segmentation of characters is not properly done.
In this paper, a new approach is proposed to solve this problem. In addition to evaluating mutual correlation between image data of each target character and candidate template, dynamic programming method (DP matching) is applied to combine segmentation and recognition of characters. Prescribed rules and knowledges related to character strings are used for matching. Proposed approach is effective for not only constant pitch strings, but also constant gap strings.
Experiments are done to show effectiveness and usefulness of this approach with respect to improvement of recognition rate under actual operational condition.
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/
