Stroke Segmentation of Reconstruct Offline Handwriting Diagram Based on Continuous Hidden Markov Model
Liming Wu, Yingmin Zhang, Yaohua Deng, Xiuchun Tang · 2009
In this paper, we experiment the capabilities of continuous density Hidden Markov Model (CHMM) to model the offline diagram sketch signals such as the electrical circuit diagram and the flowchart diagram . We attempt to imitate the online signals by extracting the offline diagram data as the time-varying coordinate sequence based on the Gradient Sharpening and Freeman code, considering that is generated by a two-level stochastic process. The underlying process governs the stroke production from a neuron-motor control point of view: go straight line, change direction line, produce a curve. A second stochastic process delivers the continuous density observed signal, which is the sequence of offline extracted points. A stroke segmentation technique based on CHMM architecture and geometric features is proposed. On a dataset of 180 hand-drawn sketches, the proposed method allows to classify correctly more than 90% of the points with respect to the connector and symbol classes.