This document contains a description of the Release 2 of the ABBYY Receipt Capture SDK 1.1.

The new Release of the product that includes:

• Official support for Corporate Travel Expense Management (CEM) scenario in USA.
• Fine purchase type classification for CEM in France by new field: Guest Count.
• Technical preview of automatic country detection function.

Key product features

The product officially supports retail tape receipts from the US and Corporate Travel Expense Management (CEM) scenario for USA and France. All other countries support is for demo purposes only.

Key features:

• Confidence and IsSuspicious properties for Vendor, Date, Total, and Time fields.
• SKU, City, Administrative region, and ZIP fields.
• Region property of a filed for easy filed coordinates access.
• Support for input in PDF format.
• Improved accuracy for Total, Date, and Line Items fields on Japanese receipts
• Additions to XML exporting parameters to support new field properties.
• XML schema update to reflect new API additions.
• Pre-trained template support for more vendors.
• Measurement unit field preview for LIs.
• Changes to the Demo sample so it displays all extracted fields and hides empty ones.
• Official support for Corporate Travel Expense Management (CEM) scenario for USA and France.
• Fine purchase type classification for CEM in France by new field: Guest Count.
• Technical preview of automatic country detection function.
• New countries in technology preview for beta-testing: Turkey, Poland.
• Improvements in technology preview by prospects' requests for UK and Australia.

Build information

• Part Number: 1358/7
• ABBYY Receipt Capture SDK 1.1 R2 for Windows
• Build Number: 1.1.173.10

Upgrading from previous versions and releases

Binary incompatibility

It is necessary to recompile a host application regardless of the SDK version used previously. Product doesn’t work with RC1.0 license. New licenses for RC1.1 are required.

New and updated features

Release 1

Confidence of extracted value for Vendor, Date, Total, and Time fields

New properties for the Vendor, Date, Total, and Time fields helping with routing of an extracted value were added: if the product is confident with it there is no need to send the value to a human for verification.

For your convenience, there are:

• IReceiptVendorField::IsSuspicious returning FALSE if the processing engine is confident that the vendor has been determined correctly. It allows 3% of errors which is known level of human accuracy.
• IReceiptVendorField::Confidence returning the confidence of vendor attribution as a number from 0 to 100 which estimates the probability that the vendor is recognized correctly for this receipt. This is for those who wants to adjust its own percentage of errors that the product does.

Please remember that allowed % of errors corresponds to % of receipts found. If you allow less errors, you will receive less receipts belonging to a certain vendor. The dependency is not linear so it might be that 1% of errors corresponds to dozens % of extracted receipts.

The property IsSuspicious has been trained on so it shows the following accuracy:

Precision shows how many correct vendor name you get by filtering the returning values with IsSuspicious=FALSE.

Recall indicates how good the technology finds and extract (not missing) values with IsSuspicious=FALSE.

ZIP, administrative region, and city components for address field

In addition to the recognized text API returns components of an address:

• ZIP – postal code found in the recognized text.
• Administrative region – state/region found in the recognized text.
• City – city name found in the recognized text.

SKU property of a line item

New property and component of the IReceiptLineItem interface return the stock-keeping unit identifier (SKU) of the product detected in the recognized text.

Region property of a field

Each recognizable field now has the Region property providing an access to coordinates of a filed value on a pre-processed image of a receipt.

This property is helpful if one needs to indicate field values on a receipt image to an end-user in order to improve UX and ease the verification process.

Support for PDF file format as input

Starting from Beta 8 release RC SDK 1 supports PDF file format for input images. This receipt image format is popular for e-mail attachments.

To enable the feature, one needs a license with corresponding option switched on.

XML parameters

As RC SDK abilities in XML export grow IXmlExportParams transforms becoming more powerful.

Now instead of simple DetailsLevel parameter one can find many more as Boolean properties at root level:

This ‘family’ of parameters is going to grow more in later releases giving you flexibility in forming desired XML output.

XML schema update

XML exporting schema become wider along with new capabilities of the product. From now on you can find the following new elements and attributes in the outputted XML file:

• Vendor field sub-elements:
  • VatNumber
• Address field sub-elements:
  • City
  • ZipCode
  • AdministrativeRegion
• Line Item sub-elements:
  • SKU
  • Amount

More pre-trained templates for vendors

Comparing to R1 in this release:

• New 19 vendors are supported in the classifier;
• 17 vendors have got the template technology support;
• 11 vendors are removed from the classifier to make it more accurate.

The full list you can find in the Appendix A to this document.

Preview of line item measurement units

The release previews new IReceiptLineItem interface property for extracting product measurement unit. AmountUnits can return one of the following values:

• Unknown
• Piece
• Liter
• Milliliter
• Fluid Ounce
• Gallon
• Kilogram
• Gram
• Quintal
• Pound
• Ounce
• Meter
• Centimeter
• Foot
• Inch
• Point

CEM support for France

Corporate Travels Expense Management (CEM) needs an employee to submit a report on its expenses during a business trip. The report includes data on an expense type, date, and amount.

With the help of our customer we shaped out critical requirements to the scenario in France and deliver corresponding technology that extracts:

• Purchase type
• Date of a purchase
• Total amount of a purchase

for 'receipts' (receipts, tickets, bills, ...) from restaurants, gasoline stations, toll roads, car parking, hotels.

As it is very important for meal expense accounting the technology also extracts applied taxes from restaurant receipts.

Expanded beta-testing program

The technology preview list gets new countries:

• Turkey. LM scenario. Expanded receipt fields support for 'A101', 'Bim', and 'Şok' vendors: Cash Register ID and #, Receipt #, and Store #.
• Poland. LM scenario. Support for VAT #, Date, Time, and Total fields. Vendor field is trained to 13 names identified by a prospect:
  • VISTULA GROUP
  • TCHIBO WARSZAWA
  • SMYK
  • Sklep Original Marines
  • RESERVED
  • PBH S.A
  • H&M Hennes&Mauritz
  • GREENPOINT
  • DEICHMANN-OBUWIE
  • COSTA COFFEE
  • CARREFOUR
  • C&A
  • Bijou Brigitte
• UK. Previously announced LM scenario support has been updated with improved receipt fields extraction accuracy for ASDA and Tesco vendors.
• CEM scenario for Australia. In a course of PoC (Proof of Concept) RC SDK has support for VAT number field and improved support for Vendor Name field. The technology was trained to recognition of 25 vendor names in Australia (see the list below). The functionality seems to be not enough for MVP in CEM scenario, thus it is still in the 'technical preview' status. Further development is scheduled for this year.
  • 7-Eleven
  • ALDI
  • Australia Post
  • Big W
  • Bunnigs
  • Caltex
  • Coles
  • Coles Express
  • Dan Murphy's
  • David Jones
  • Foam Coffee Bar
  • IGA
  • Jaycar Electronics
  • JB Hi-Fi
  • Kmart
  • Lucky 7
  • Masters Home Improvement
  • McDonald's
  • Metro Petroleum
  • Officeworks
  • Spar
  • Supercheap Auto
  • United
  • Woolworths
  • Woolworths Petrol

Release 2

CEM support in USA

CEM (Corporate (Travel) Expense Management) feature helps to automate the process of submitting a travel expense report to your organization after or during a business trip. RC technology is capable of extracting required data from a document picture (scan or photo).

With the technology you can process documents proofing different purchasing types:

• Meal at restaurants
• Fuel purchases at gasoline stations
• Accommodating in hotels
• Taking a taxi

The purchasing types above form the most of a business person expenses in a travel. Of course more types are there like air/train bookings, toll road passing, etc., but their share is insignificant in reports - one can process them manually w/o noticeable burden.

The following data is extracted automatically from the documents:

• Document/purchase type
• Date/Check-in date/Check-out date
• Total
• Line items: date, description, total (applicable for hotel bills)

Note: at the moment each page of a multipage hotel bill is processed individually and a RC SDK user is responsible for combining results from all pages.

CEM support in France update

The technology is capable of assigning the purchase type 'Restaurant' to receipts from restaurants. This is not enough for the financial law in France: a company needs to attribute a receipt either to a meal (employee nutrition) or to a customer/partner invitation (business meetings).

There is the simple workaround: the technology marks a receipt as a restaurant receipt and extracts a number of guests printed on a receipt, then a customer makes a final decision by the number of guests - 1 guest means 'meal', 2 and more guests mean 'invitation'.

Loyalty Management support in USA update

Loyalty Management scenario gets continuous improvements. In this release:

• JCPenney vendor has got 'templated' technology support which means higher extraction accuracy for all fields and especially for line items and its components. In addition the technology extract 'Cashier' field from receipts of this vendor. The field is required for de-duplication process and fraud protection in LM scenario.
• XML export presents <streetAddress> element containing Receipt::AddressField::RecognizedText value for the sake of clarity. Previously XML element <address> represented both a full vendor address and a street vendor address depending on the technology capabilities for different countries and/or document types. Now a full address goes to <fullAddress>, a street address goes to <address> elements respectively.

Vendor name classifier update for USA

Vendor Name field employs several technologies in order to extract correct value from a receipt. They are:

1. Vendor Logo classifier. It knows some picture specific to particular vendors and can recognize those pictures in a receipt. This is very accurate classifier: if it fails then most probably a receipt belongs to unknown vendor.
2. Receipt Text classifier. It works with full-text OCR result and knows specific words (keywords) for each known vendor. This classifier works of Logo Classifier fails. Moreover, a customer is allowed to append pre-trained classifier or train own one.
3. 'Templated' classifier. It knows a vendor receipt layout: static text, keywords, and field values relative positions. Receipt layouts are pretty unique, but creating them is time consuming, thus the product includes templates for the most important vendors. This classifier works last in a row and adjusts decisions of Logo and Text classifiers.
4. Generic technology finder. This is an algorithm guessing a vendor name out of full-text OCR result by common keywords, position on a receipt, font size, similarity to an URL at the bottom of a receipt, etc. This technology is much less accurate then all above, that is why it works when all above failed.

In this release Text classifier has got 3x times bigger coverage: 337 known vendors. The name list is below.

Notes: 

• The updated classifier can't be extended with new names by now, this is scheduled for the next release. If you need a custom text classifier or want to append pre-trained one please use legacy version of Text classifier which is available provided you set IReceiptSynthesisParams::UseAdvancedVendorClassifier = false.
• The updated classifier works for USA only and it bypasses 'Templated' classifier. Balanced superposition of Logo, updated Text, 'Templated' classifiers is scheduled for the next release

Customer support

The ABBYY SDK Support team is ready to help you. Please refer to the contact information and hours below.

Contacts

• ABBYY Technology Portal: https://abbyy.technology
• Customer Support Management (CSM) Portal: www.abbyy.com/csm
• Developer Support for Receipt Capture SDK Beta 9 for Windows:
  Email: barbara.gross@abbyyusa.com
  Office Hours: Monday – Friday from 9AM to 6PM PST

Information required

When opening a support case or contacting support, please be prepared to provide the following information:

• Description of the issue
• Sample images for testing
• Run and send Ainfo Report from the Bin/Support folder of the installation directory
• Full error messages that have occurred
• Any additional information you feel may be helpful for the investigation

The information above will assist the ABBYY Support team in investigating your issue and in providing a prompt response.

Appendix A. The list of supported US vendors

Added support is bold. Removed items are stroke out.