A Kalman approach for stroke order recovering from off-line handwriting

Non-constrained handwriting recognition is still faced with very difficult problems. However, on-line recognition methods exhibit better results than offline methods which lose all temporal information. The aim of the work presented here is to recover the strokes ordering from static 2-D images as it is inherently available from on-line systems. The approach presented here is innovative, first conversely to most other ones, it uses extensively the gray-level information, secondly because of the use of the Kalman filtering approach in the prediction of the writing strokes trajectory.