Multi-task Layout Analysis for Historical Handwritten Documents Using Fully Convolutional Networks

Layout analysis is a fundamental process in document image analysis and understanding. It contains three key sub-processes which are page segmentation, text line segmentation and baseline detection. In this paper, we propose a multi-task layout analysis method that uses a single FCN model to solve the above three problems simultaneously. In our work, a multi-task FCN is trained to segment the document image into different regions (background, main text, comment and decoration), circle the contour of text lines and detect the centerlines of text lines by classifying pixels into different categories. By supervised learning on document images with pixel-wise labeled, the FCN can extract discriminative features and perform pixel-wise classification accurately. Based on the above results, text lines can be segmented and the baseline of each text line can be determined. After that, post-processing steps are taken to reduce noises, correct wrong segmentations and produce the final results. Experimental results on the public dataset DIVA-HisDB [Simistira et al., 2016] containing challenging medieval manuscripts demonstrate the effectiveness and superiority of the proposed method