Cheque Information Processing and Validation using Deep Learning

Understanding document images, such as invoices, is a difficult task. It necessitates reading the text as well as comprehending the document’s general structure. Conventional techniques first extract text using Optical Character Recognition (OCR), yet this method has drawbacks. In addition to being computationally costly, OCR can have trouble converting between various languages and document formats and introduce errors that can compound throughout the analysis process.Donut is an OCR-free VDU (Visual Document Understanding) model that challenges this paradigm. It does not even bother with the text extraction stage; it analyzes the document image straight. This creative strategy produces notable advancements. Donut performs at the cutting edge, which means it is quick and precise. This is a significant development in the realm of visual document understanding since it enables Donut to handle a greater variety of languages and document types. The model has been tested using BLEU scores and ROUGE scores over 3 parameters namely – payee name, amount in words and amount in figure. The model has also been tested over variety of sample sizes ranging from 50 to 1000.