Contextual text/non-text stroke classification in online handwritten notes with conditional random fields

作者：

Highlights：

• We present a comprehensive study on the exploitation of context for text/non-text separation in free form online documents.

• The problem is formulated as a Conditional Random Field for labeling the document as a graph of strokes.

• Different sources of context (spatial, temporal…) are modeled as template cliques and combined in the CRF.

• Results on a public database determine new state-of-the-art performance.

摘要

Highlights•We present a comprehensive study on the exploitation of context for text/non-text separation in free form online documents.•The problem is formulated as a Conditional Random Field for labeling the document as a graph of strokes.•Different sources of context (spatial, temporal…) are modeled as template cliques and combined in the CRF.•Results on a public database determine new state-of-the-art performance.

论文关键词：Online handwritten documents,Structured prediction,Stroke classification,Conditional random fields,Text/non-text separation

论文评审过程：Available online 2 May 2013.

论文官网地址：https://doi.org/10.1016/j.patcog.2013.04.017