Exploiting ruling line artifacts in writer identification

In this paper, we address the writer identification problem for noisy handwritten documents written on a substrate of pre-printed ruling lines. Instead of attempting to remove rulings and to recover broken strokes, we incorporate rulings to help with the identification task through the use of new displacement features. Experiments involving 61 writers and 4,890 handwritten text lines show that our technique is effective, with a relative 10% performance gain over the baseline system which attempts to remove ruling lines and recover broken strokes.