A Maximum-Likelihood Approach to Symbolic Indirect Correlation

Abstract

Abstract

Symbolic indirect correlation improves text recognition

“…Polygram matches between the sample in question and a set of labeled reference words are used to derive the query likelihood for each class, P(query|class). 3,7 Then, like character-/sub-character recognition schemes, SIC builds a lexical transcript by imposing grammatical constraints on the assigned polygrams. Unlike these methods, however, unconstrained partial word matching does not rely on the error-prone segmentation of the query into letters or strokes.…”

Multi-character Field Recognition for Arabic and Chinese Handwriting

“…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]. A maximumlikelihood approach [10] avoids this computational bottleneck in the second-level matching. Since this method seems promising for Arabic recognition, we describe in some detail how we build candidate solutions to the query; interested readers will find full technical details in [10,11].…”

“…However, there are many spurious matches as well. Our second-level matching process, described in [10] is shown to be robust against a large number of spurious matches but at the expense of increased computation time. Therefore, we plan to explore a post-processing approach to eliminate some bad matches.…”

“…A maximumlikelihood approach [10] avoids this computational bottleneck in the second-level matching. Since this method seems promising for Arabic recognition, we describe in some detail how we build candidate solutions to the query; interested readers will find full technical details in [10,11].…”