On-Line Recognition of Handwritten Whiteboard Notes: A Novel Approach for Script Line Identification And Normalization

In this work we use a previously published approach for script line identification of handwritten whiteboard notes in order to perform skew correction and size normalization of the script trajectory. Arbitrary assignments of sample points to certain script lines are hypothesized and described in a trellis. The normalization is performed by equalizing the script lines and warping the script trajectory accordingly. In an experimental section we show that the novel normalization achieves a relative improvement of r = 1.6 % in character level accuracy and r = 1.4 % in word level accuracy compared to a system using standard normalization.