PAGE IDENTIFICATION METHODS

PAGE IDENTIFICATION METHODS

Taskmaster supports several methods for page identification,including but not limited to:

• Fingerprint matching


•  Structure-based identification


•  Text matching


•  Manual page identification

Additionally, if your application supports only a single page type, you can simply

assign a static page type to all incoming pages. This section provides an overview

of these page identification methods.

FINGERPRINT MATCHING
With fingerprint matching, Taskmaster generates a “fingerprint” that describes

each incoming page. The fingerprint can include information about the relative

densities of different regions of the page or the location of text on the page.

Taskmaster then compares the new fingerprint to a library of fingerprints for

known page types. When it finds a match it assigns the corresponding page type.



In the example above, the incoming page matches the TopSuite room receipt.

Taskmaster assigns it the type “Room_Receipt” and records the ID of the

matching fingerprint in the runtime batch hierarchy. The match will not be

exact since the data on the page will most likely be different, but we‟re looking

for the best match.

SELECTING THE FINGERPRINT CREATION MODE
Taskmaster provides two primary methods for generating page fingerprints:

 Image analysis: This scans the page image to identify the composite

“blackness” of different regions of the page. This method provides fast

page identification, but requires that you perform recognition later.

Full page recognition: This performs optical character recognition to

identify the locations of text within the page. This method takes longer,

especially with pages that include handwritten text, but cuts time from

subsequent workflow tasks since the full page recognition results are

available for use.

Both of these methods write the resulting information to a“.cco” file

that‟s stored with the original .tif image file in the application‟s“fingerprint” folder.

USING IMAGE ANALYSIS
Image analysis uses a pixel-based algorithm to generate a fingerprint (.cco) file

that represents the relative blackness of different regions of the page.

The AnalyzeImage action in the “Recog_Shared” actions library performs image

analysis on an image file.

USING FULL PAGE RECOGNITION
Full page recognition, as its name suggests, uses the text and location of text on

the page to generate the fingerprint (.cco) file. Taskmaster includes three optical

character recognition (OCR) engines, plus one intelligent character recognition

(ICR) engine that you can use to perform full page recognition:

•  OCR_a: ABBYY FineReader OCR engine


•  OCR_s: Nuance (formerly ScanSoft) OmniPage OCR engine


•  OCR_sr: Newer implementation of the Nuance OmniPage OCR engine


•  ICR_c: Open Text RecoStar ICR engine

Additional ICR engines are also available as options. As a general rule, the OCR

engines work well with machine printed text, whereas the ICR engine works

well with hand printed as well as machine printed text.Taskmaster include

actions libraries for each recognition engine (ocr_a, OCR_s, ocr_sr, and icr_c).

Each   library includes its own version of the full page recognition action.

STRUCTURE-BASED PAGE IDENTIFICATION

Structure-based identification uses the position of a page within the batch

to determine its type. If your application handles only one page type, or if

the document structure is consistent (for example, all documents are two

pages with a main page and a trailing page), you can assign page types based

on position. You can do this using the SetPageType action.If a batch contains

documents of varying length, you can use separator pages between

documents.For an example that uses barcoded separators, look at the

Taskmaster Accounts Payable (APT) foundation application included with

Taskmaster.When you identify a page using structure-based identification,

the page is not matched to a fingerprint, and so there are no recognition

zones for your application to locate data during recognition. You can

design your application to locate data fields using keyword identification

or pattern matching techniques that do not relyon recognition zones.

We‟ll do this in a later chapter in this guide.

TEXT MATCHING
To perform page identification using text matching, you must first perform

full page recognition. You can then search the recognition results for a

string that‟s unique to each page type.In the example below, the first function

performs full page recognition and looks for the string “Car” on the current

page.If it finds it, it assigns the page type “Rental_Agreement”; if it doesn‟t the

function fails and the second function looks for the string “Flight.” If it finds it,

it assigns the page type “Air_Ticket”; if it doesn‟t the function fails and the

third function looks for the string “Room.” If it finds it, it assigns the page type

“Room_Receipt”; if it doesn‟t the page remains with the page type “Other.”



As with the structure-based techniques, when you identify a page using text

matching, the page is not matched to a fingerprint, so you‟ll have to use a

recognition technique that does not rely on recognition zones. We‟ll cover

this later in the chapter on text matching.

MANUAL PAGE IDENTIFICATION
The page identification techniques described so far all identify pages

automatically. It‟s also possible to configure your application to display

unrecognized pages to an operator for manual identification. You can do

this at scan time or during verification; however these techniques are

beyond the scope of this guide.