Visual-Linguistic Methods for Receipt Field Recognition

Autor:	Nimrod Morag, Rinon Gal, Roy Shilkrot
Rok vydání:	2019
Předmět:	Receipt Network architecture Computer science Embedding Variance (accounting) Optical character recognition computer.software_genre computer Field (computer science) Linguistics Task (project management)
Zdroj:	Computer Vision – ACCV 2018 ISBN: 9783030208899 ACCV (2)
DOI:	10.1007/978-3-030-20890-5_35
Popis:	Receipts are crucial for many businesses’ operation, where expenses are tracked meticulously. Receipt documents are often scanned into images, digitized and analyzed before the information is streamed into institutional financial applications. The precise extraction of expense data from receipt images is a difficult task owed to the high variance in fonts and layouts, the frailty of the print paper, unstructured scanning environments and an immeasurable amount of domains. We propose a method that combines visual and linguistic features for automatic information retrieval from receipt images using deep network architectures, which outperforms existing approaches. Our Skip-Rect Embedding (SRE) descriptor is demonstrated in two canonical applications for receipt information retrieval: field extraction and Optical Character Recognition (OCR) error enhancement.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::b1ea0de0302655da611f450f04d22a5d https://doi.org/10.1007/978-3-030-20890-5_35 Zobrazit plný text záznamu