Analyzing the Potential of Layout Analysis Systems for the Task of Shopping Receipts Analysis on ReceiptDB Dataset

Muhammad Shoaib Ahmed Siddiqui, Soumen Pramanik, Pervaiz Khan, Andreas Dengel, Sheraz Ahmed



Shopping receipts are an important form of document which is ubiquitous as a proof of transaction all over the world. Despite recent advancements in the domain of document analysis, analysis of documents with complex layouts like receipts is still not a reality. In order to further investigate the efficacy of state-of-the-art layout analysis systems in the document processing community for the task of shopping receipt analysis, we curated a custom shopping receipts dataset comprising of 539 receipts collected from over 10 different supermarkets in Germany named as ReceiptDB. The dataset is densely labeled with the most important information which includes information regarding the row, product name, price, description, header, footer, logo, total price and total price text. Furthermore, in order to establish a baseline, we employed a state-of-the-art document analysis system powered by deformable FPN. It is evident from the obtained results that solving the problem of shopping receipt analysis will require significant efforts from the document analysis community in combination with the advances in deep learning literature.

Deep_Receipt_Latest_(1).pdf (pdf, 13 MB )

German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz