Automatic reading and interpretation of paper invoices : ADC invoice

Sammanfattning: Manually typing long numbers on paper invoices is tedious and timeconsuming work. The task of typing all the fields of an invoice into a text editor was given to two subjects working with bookkeeping, and the average time consumed was measured to be five minutes. The time and cost spent on manual typing will accumulate for companies that receive a lot of invoices. Swedbank along with other banks, have addressed this issue with a mobile application that reads and interprets the numbers on an invoice using the built in camera. This solution is directed to the public and the extracted information cannot be imported into bookkeeping software. A standalone software for digital reading and interpretation of scanned invoices is our solution for companies in regards to this issue. There is already technology available that will interpret written text. These techniques were applied in our work, but the focus of this project was to implement algorithms for finding the location for a specific number and resolve what bookkeeping terms the number references e.g. OCR, IBAN numbers, etc. This is hampered by the missing of a general invoice layout. 52% of all the sought information was extracted correctly and almost 95% of the bookkeeping details that are changed most frequently on invoices from the same service providers were extracted correctly. The average time it takes for the application to extract the vital data is 30-40 seconds.