Receipt Transcription
Currently, Taggun can recognise and extract the following properties from a receipt:
Total amount
Tax amount
Date
Merchant name
Merchant address
Merchant type
Line amounts (a list of amounts detected for each line item)
Line description (a list of description detected for each line item) *alpha
Amounts (a list of all extractable amounts found on a receipt)
Numbers (a list of reference numbers found on a receipt)
Invoice number *alpha
IBAN bank account number *alpha
Taggun supports the following image and file types:
JPEG
PDF
PNG8
PNG24
GIF
Taggun provides different levels of support for receipts from different countries.
Category 1 has a high accuracy rate. Taggun actively measures and improves the accuracy rate for receipts from these countries:
USA
Canada (English & French)
United Kingdom
Australia
New Zealand
France
Belgium (Dutch & French)
Category 2 has a medium to high accuracy rate. Taggun provides support for receipts from these countries:
Israel (Hebrew)
Germany
Japan
China
Taiwan
Singapore
Malaysia
Brazil
India
Spain
Mexico
Sweden
Czech Republic
Finland
Poland
Norway
Category 3 is recognised to have a low accuracy rate. Support is limited to our ability to assist:
Thailand
Arabic
Russia
Hungary
Handwritten receipts
Taggun calculates the confidence level for each property. This provides a "proxy" accuracy level for each property. Also, an aggregated confidence level for all properties of the receipt is provided at the root level of the result. Maximum confidence level is 0.99 Minimum confidence level is 0.
Taggun uses Google Places to extract and validate the recognised merchant name and address of the receipt. It biases the result to the closest distance of the bias location (the geolocation of user's IP address or caller's IP Address).
To improve the accuracy of merchant name for your account, you can email us a list of the merchant names in your system. Or you can add a new merchant name to your account with the /api/account/v1/merchantname/add endpoint.
What happens when the original location of the receipt is not the same as the bias location of the user or caller? E.g. A user in Australia has scanned receipts from overseas trip in the USA. Think of IP Address geolocation as a mere "suggestion" to influence the result. It is not deterministic. Taggun algorithm is robust enough to extract information without any information of the IP Address.
When possible, it is recommended to include the user's IP Address to lookup for the bias location of the receipt. Include ipAddress request parameter to improve the accuracy of receipt transcription. Taggun uses GeoLite2 data created by MaxMind, available from http://www.maxmind.com.
Caller's IP address is the IP address of the server that makes the API request. When the user's IP address is not available, Taggun uses the caller's IP address to lookup for the bias location of the receipt.
Taggun recognises dates of any formats. Bias location is used when there is an ambiguity between little-endian (DD-MM-YYYY) and middle-endian (MM-DD-YYYY). For example: a request with the bias location of New York, USA will recognise 07-12-2017 as 12th of July. But the same request with the bias location of Auckland, New Zealand will recognise that as 7th of December.
Taggun recognises both formats decimal point(.) and decimal mark(,) to extract the correct amount.
