Automatic data labeling for document image segmentation using deep neural networks

The article proposes a new method for automatic data annotation solving the problem of document image segmentation using deep object detection neural networks. format marked PDF files is considered as initial markup. peculiarity this that it includes hidden marks describe logical and physical structure document. To extract them, tool has been developed simulates operation stack-based printing machine according to specification. For each page document, an are generated in PASCAL VOC format. classes coordinates bounding boxes calculated during interpretation labeled file based on labels. test method, collection up was formed from which images pages annotations three (text, table, figure) were automatically obtained. Based these data, network EfficientDet D2 architecture trained. model tested manually same domain, confirmed effectiveness applied problems.