A Deep Convolutional Encoder-Decoder Network for Page Segmentation of Historical Handwritten Documents Into Text Zones

Recent research activity for page segmentation and pixel-labeling problems focuses strongly on deep Neural Network architectures. In this paper, we present a Convolutional Encoder-Decoder based method for the segmentation of historical handwritten images into distinct text zones. This is achieved by labeling each pixel of the image to one of the predefined classes (main body, comments, decorations, periphery, background). Traditional methods make use of prior knowledge of documents and rely on data-oriented features and experimental rules. We propose a method using Convolutional Encoder-Decoder pairs and we show that deep architectures fit properly to our problem. Experiments on different public datasets demonstrate the effectiveness of the proposed method that outperforms previous techniques in many cases.