DCPL

This study constructs a new dataset named Documents with Complex Page Layouts (DCPL). The dataset consists of grayscale scanned images of real documents, covering various types such as journals, magazines, and academic papers. It features diverse layout styles and heterogeneous content structures. The documents indude formulas, tables,illustrations, and multlingual text, with languages including but not limited toEnglish, Chinese, and Russian.
The page layouts exhibit complex and highly variable structures, incuding two-column formats,embedded images and text, mixed text -imagearangements, and irregular paragraph layouts, effectively representing a wide range of document styles encountered in real-world scenarios.The entire DCPL dataset contains 900 document pages.