Combined Segmentation and Recognition of Online Handwritten Diagrams with High Order Markov Random Field

Chengcheng Wang 1 Harold Mouchère 2 Christian Viard-Gaudin 2 Lianwen Jin 1
1 School of Electronic and Information Engineering
2 IRCCyN - Institut de Recherche en Communications et en Cybernétique de Nantes
Abstract : In this research we focus on the recognition of online handwritten diagrams, which is widely applied in note recording. Handwritten diagram is a kind of 2D language that spans in the plane, consisting of symbols and structures. Previous researches in the field of 2D language symbol recognition involved a complicated step such as symbol hypothesis generation and grammar description. By building a three order Markov random field on the stroke level, we not only labeled strokes but stroke relationships so that we could complete symbol grouping and symbol recognition simultaneously. The potential function in our Markov random field is log-linear model so that it is fully data-driven, and we trained it using max-margin method. We tested our method on two public handwritten diagram datasets and experiment showed that our symbol recognition method’s performance has reached state-of-the-art.
Keywords : High-order Markov random field Max-margin Markov random field Online handwritten diagram recognition Global recognition
Document type :
Conference papers
Complete list of metadatas
https://hal.archives-ouvertes.fr/hal-01374389
Contributor : Harold Mouchère <>
Submitted on : Friday, September 30, 2016 - 1:21:47 PM
Last modification on : Tuesday, October 15, 2019 - 11:00:04 AM

Identifiers

  • HAL Id : hal-01374389, version 1

Collections

EC-NANTES | IRCCYN | UNAM | UNIV-NANTES | IVC-IPI

Citation

Chengcheng Wang, Harold Mouchère, Christian Viard-Gaudin, Lianwen Jin. Combined Segmentation and Recognition of Online Handwritten Diagrams with High Order Markov Random Field. International Conference on Frontiers in Handwriting Recognition (ICFHR), Oct 2016, Shenzhen, China. ⟨hal-01374389⟩

Export

BibTeX TEI DC DCterms EndNote

Share

Metrics

Record views

313

Contact

support.ccsd.cnrs.fr
hal.support@ccsd.cnrs.fr
Logo CCSD