Dynamic Time Warping for Chinese calligraphic character matching and recognizing

Historical Chinese calligraphy is now being scanned by ongoing project of Universal Digital Library to enable universal access. 483 Calligraphic page images are segmented into 13,351 individual characters, followed by character feature extraction. When an image query submits, a short list of candidates from the database is selected according to simple features of stroke transect. Then the selected candidates are ranked by shape matching using two-dimensional Dynamic Time Warping (DTW). After that, similar calligraphic characters are retrieved and ranked, then the query is recognized as the label of the most similar retrieved characters.

► 483 Calligraphic page images are segmented into 13,351 individual characters, followed by character feature extraction.
► First, a shorter candidate list is select from the database by pruning based on features of stroke transect.
► Then the selected candidates are ranked by developing extended two-dimensional Dynamic Time Warping (DTW).
► Finally, the query is recognized as the label of most similar retrieved characters.