Towards unsupervised learning and graphical representation for on-line handwriting script

To train cursive script recognition system, a large labeled database at different levels (grapheme, character or word) is required. Nevertheless, manual segmentation and labeling are tedious tasks. To reduce the human workload, we are motivated to automate the annotation process. Considering online handwriting problems and the Arabic script characteristics, we discuss the implementation of word recognition system based on unsupervised approaches. Word segmentation is performed into strokes as written by the writers. Then, Agglomerative Hierarchical Clustering is used to produce a Codebook with one stroke per class. This Codebook is labeled manually. Using spatial relations, we introduce a new representation for online Arabic handwriting which is graphical representation.