Kalman filter contributions towards document segmentation

One way of tackling document analysis is to split the document into different layers and then to find a segmentation for each one. The straight line is often one of the basic elements of the layers making up a document. The detection of these segments is often disturbed by the superposition of different layers in the same area. Therefore we developed a segment extractor, which processes masked areas, discontinuities and even curvatures. After showing in our previous work, the interest of such a system in binary image analysis, our aim was to show the method could be generalized to grey level images, by taking into account pixel luminosity as one of the parameters. The Kalman filter method combined with an adequate control scheme, allows the pooling of local observations, when their position, width and luminosity are coherent. The proposed model gives an interesting scheme to manage drawings masking each other (eg. cross-lines).