Abstract
This paper proposes a new systematic approach of analyzing and recognizing difficult Braille characters. The approach is based on integrating image processing and fuzzy reasoning, where fuzzy reasoning played key role in formulating the classification algorithm, and image processing in identifying effective features to be used in the fuzzy rule evaluation. We introduced five features(observable) on the amounts of bright-or-dark-pixels (T), highlighted-pixels (H), shaded-pixels (S), pixels included in area which boundary is defined by sharp change in the brightness (C), and pixels included in boundary area (E), and to each of which membership functions are introduced to represent deformed Braille-dot sets of "worn-down" "stained", "cracked", "holed", and "normal", that are mapped to membership functions defined to the feature variables space respctively. The fuzzy clustering analysis was used to find the five feature parameters, and the membership functions were trained to 5600 teacher data points. The new recognition approach showed very high accuracy of 99.99%. Since our approach recognizes also the types of deformed Braille character, it fines quality level of legacy books easily for recommending whether refreshing should be made or not to refresh them.
