Complex handwritten page segmentation using contextual models

In this paper we address the problem of segmenting complex handwritten pages such as novelist drafts or authorial manuscripts. We propose to use stochastic and contextual models in order to cope with local spatial variability, and to take into account some prior knowledge about the global structure of the document image. The models we propose to use are Markov Random Field models. After a formal description of the theoretical framework of Markov Random Fields and the principles of image segmentation using such models, we describe the implementation of our model and the proposed segmentation method. Then we discuss the results obtained with this approach on the drafts of the French novelist Gustave Flaubert, for different segmentation tasks. In conclusion, an extension of this work towards the use of discriminative models is discussed