|
|
 |
|
Legacy Forms
|
|
|
|
|
|
|
For our purposes, a legacy form is a preprined form which was not prepared or optimised by the
author for automatic data capture from scanned images using recognition technologies (ICR).
Usually, the ability to process legacy forms is an interim step until optimised forms are in
circulation.
|
|
|
|
|
|
|
Typical Legacy Form Data Capture Problems
|
|
|
|
|
|
|
- Data boxes too small - filled data extends beyond box
- Intruding data from neighouring boxes
|
|
- Unconstrained data fields rather than semi-constrained
|
|
- Pre-printed text merging with filled data
|
|
|
|
|
|
|
|
|
|
|
|
Non-field Based Processing
|
|
|
|
|
|
|
Unlike optimised form processing, legacy form processing solutions must be capable of searching for
hand-filled data in the surrounding region of the data box and then be able to isolate the data in
relation to any surrounding data.
In the example below, Palmares's
ADCS Recognition Server was able to isolate the telephone number
then segment the characters. The recognition character results and their confidence are provided as
information.
|
|
|
|
|
|
|
An Example Project Task
|
|
|
|
|
|
|
An example project task:
- analyse 600 data entry boxes for hand filled data and then return the appropriate article number.
- document layout redesigned every three weeks.
With traditional Field Based processing, as shown in the example above, the "1" straddles
the boundary of two adjacent fields producing two substitute errors. This could be very difficult
to detect at the verification phase.
The relational non-field based method is not restricted to looking in predefined
fields like traditional solutions but dynamically searches larger zones for the presence of data
fragments.
Once fragments have been detected they are analysed relationally with respect to the adjacent part
numbers to determine whether the hand fragments potentially form part of a valid character or word.
|
|
|
|
|
|
|
Relation Verification (patents pending)
|
|
|
|
|
|
|
The ability to display the recognition results to the verifier operator in such a way as to trap
any potential substitutes is important in legacy form recognition.
In our relational non-field based
project example, the
Palmares ADCS Verifier provides the operator with
both the relational results and the oportunity to correct the relationship in the case of an error.
The relational non-field based processing provided by the Palmares ADCS solution does tend to
improve the segmentation of characters and words compared with traditional field based methods.
|
|
|
|
|
|
