A new approach for centerline extraction in handwritten strokes: an application to the constitution of a code book

To obtain a good decomposition of handwritten strokes in manuscript images, the extraction of the medial axis represents a very important step. In this paper, we propose an automatic approach for the extraction of the median axis directly from the gray level images. This approach is based on the orientation and the stroke thickness analysis at different points of the handwritten shapes. This method is robust to degradations of the strokes, background noise and ink irregularities in the support paper. Then, we propose to consider all graphemes and to classify them from the analysis of median axis points. We use this decomposition to produce an identification scheme from various handwriting styles. This work is supported by the ANR GRAPHEM project.