Recovering writing traces in off-line handwriting recognition: using a global optimization technique

Off-line handwriting recognition is constrained by the lack of time information. This paper describes an approach to derive sequence information from a written word. The underlying representation of the word used is a graph with nodes and edges resulting from a skeletonization process. The writing path is reconstructed by assuming that the writer has used a path with least curvature, i.e. reconstruction is realized as a global optimization process, which consists of solving a travelling salesman problem.