Why an open software solution?
When a post starts moving from manual to automated mail processing, it usually purchases letter sorting machines with embedded automatic address recognition and video coding. This works well until there is a need for additional letter, flats, or parcels machines, or because some older machines require updates. This raises the problem of coexistence between different types or versions of machines. Posts want to avoid being bound to a single machine provider without facing incompatible recognition solutions, making any extension complex and costly as well as having to train its keyers on one or more incompatible video coding systems.
To address this concern many posts started integrating a single recognition solution that communicated with their different machines. This covered part of the problem but leaves the post bound to the software solution provider. When looking to increase recognition performance or developing a new service, the post is not able to take advantage of open competition and purchase the best package in the marketplace, without expensive integration and deployment costs involving the original solution provider.
A paradigm shift for some posts, truly open systems are providing state of the art technologies that are delivering on automation targets across all automation platforms, while providing posts the ability to maintain, enhance, or add-on best-of-breed solutions.
The CEN open model
The European Committee for Standardization (CEN) has defined an open model with interfaces that enable an open architecture. The recognition framework is built around an “Image Controller” which receives images from any mail processing machine and dispatches them for processing to “Enrichment Devices”. The Image Controller supports two main interfaces, one with the Enrichment Devices and one with the machines. Enforcing a standard interface with the machines provides the independence between them and the recognition software, yet the openness of the recognition software is provided by the standard interface to the Enrichment Devices, which allows the posts to evolve their solution by integrating additional Enrichment Devices without impact to machines processing the mail.
Seamless integration of an OCR
AddressVision Inc. (AVI is a Bull company) mail automation solutions comply with the CEN open model and interfaces. In order to illustrate this, we will now examine how a post using the AVI solution may improve its recognition performance by adding an OCR without requiring additional AVI intervention. The objective is not to process rejects through a “secondary” OCR rather taking full advantage of the strengths of an additional OCR.
AVI’s Image Controller includes a Configurable Workflow Management, which enables a rule driven sequencing of Enrichment Devices, based on system capacity, attributes of the image and confidence in current results. These rules are written in an interpretative language and integrators from the post modify them, in order to invoke the additional OCR under the exact circumstances selected. In addition, each new OCR takes advantage of mail batch attributes and of partial results of previously invoked OCRs using the CEN concept of “pre-knowledge”. For instance, if one Enrichment Device is dedicated to Address Block Location, its results will be transmitted to the additional OCR as “pre-knowledge”.
In instances when the additional OCR gives a result different from that of existing OCR’s, the Voting System selects the correct result. AVI Arbitration or Voting System is independently trained and tuned for each OCR, using an inductive machine learning technique. As shown in the figure below, the Arbitration trainer performs an analysis of the OCR’s capabilities, based on the assignment results from a truthed image set. The output of this analysis is a decision tree for use by the Arbitration device to select an assignment result between those issued by the different OCR’s. This training process does not imply any modification of the Arbitration software and may also be performed by the Post.
This example depicts how a truly open solution moves the control of mail automation from the provider to the post. An open solution allows recognition improvements or implementation of new services just by adding the best of breed solution into the existing framework, without being bound to any provider.