Coarse Writing-Style Clustering Based on Simple Stroke-Related Features
L.G. Vuurpijl, Lambert Schomaker · 1996
Two methods are presented for the automatic detection of generic writing styles like e.g. mixed, cursive and handprint. Based on a set of handwritten words, three features are determined: a cursivity index c, which indicates the tendency of a writer to write cursive, and two distance measures d c and d h . The distance measures represents the distance between the stroke feature vectors in the input handwriting and the strokes contained in two style-specific Kohonen Self-Organizing Maps (SOM). One SOM is tuned for the writing style handprint, and the other for cursive. The first method uses some linear decision criteria based on the feature vector fc; d c ; d h g, for the classification of one of the three writing styles. The second method uses nonlinear decision boundaries found via agglomerative hierarchical clustering of the three-dimensional feature vectors. Using the second method, several more distinctive writing style classifications are proposed. 1. Introduction At the NICI, s...