A nonparametric classifier for unsegmented text

Abstract

Symbolic Indirect Correlation (SIC) is a new classification method for unsegmented patterns. SIC requires two levels of comparisons. First, the feature sequences from an unknown query signal and a known multi-pattern reference signal are matched. Then, the order of the matched features is compared with the order of matches between every lexicon symbolstring and the reference string in the lexical domain. The query is classified according to the best matching lexicon string in the second comparison. Accuracy increases as classified feature-and-symbol strings are added to the reference string.

A Maximum-Likelihood Approach to Symbolic Indirect Correlation

“…In [7] and [8], we introduced the SIC classifier with a representation based on bipartite graphs and showed with simulations that at significant noise levels (40% spurious and 40% missed matches) the SIC classifier gives very high accuracy (97% for 50 word lexicons). A significant advantage of SIC over several existing classifiers is that it does not need feature-level samples from every class.…”

“…For the example of Table 2, the likelihood of assignment be-n-eath-ath is the product of all the elements in Table 2, i.e., 0.15 2 ×0.06×0.85×0.94 8 . The first branch assigned segment S 4 to th instead of ath, so the conditional probability for these was p 1|1 =0.15 instead of p 1|0 =0.06.…”

“…Furthermore, SIC can compensate for noisy features or inaccurate feature matching by increasing the length of the reference set. We introduced SIC in [7] and [8] with a representation based on ordered bipartite graphs and established its advantages on simulations with a significant amount of noise. In the presence of excessive noise, however, the graph-theoretic approach to SIC may need an unreasonably large reference set [9], [10].…”

Symbolic indirect correlation improves text recognition

“…We have formulated both a graph-theoretic 3,5,6 and a probabilistic 5 approach to perform the second-level comparison. Here, we discuss the probabilistic method.…”

Multi-character Field Recognition for Arabic and Chinese Handwriting

“…We introduced SIC in [6,7] with a representation based on ordered bipartite graphs and established its advantages through simulations with a significant amount of noise. Later investigations showed that in the presence of excessive noise, the sub-graph isomorphism based approach to the second-level matching requires an unreasonably large reference set [8,9].…”