Branch Image Capture

The first evolutionary phase to distributed capture is called branch image capture (BIC). Referenced as back-counter capture or capture on a deferred workstation, BIC was the first step toward minimizing courier expenses by transmitting captured images from each branch to a centralized location. Certain financial institutions have implemented 100% branch image capture, while others are more selective. A selective deployment is also known as “popcorn” installations, where certain remote branches are chosen due to distance to the main or regional processing centers.

Check recognition must be flexible in architecture to meet the varying needs in branch image capture. There are two common recognition implementations for this environment: central configurations or “thick client” where the recognition engine is located on the BIC computer. Each has pros and cons depending on the item processing application driving the scanners.

Centralized Configuration

In this environment, the item processing application must be able to handle many scanners simultaneously feeding images. This process can be handled through various technology approaches. When the item processing application opens several “threads”, this means that they are sending multiple sets of images to the recognition engine simultaneously. If the item processing application uses a database approach, they can gather images in bulk and pass them in for high-speed processing in a more synchronous interface. Both options are supported well with the Orbograph recognition solutions. Benefits to this approach is a single location for reporting, item definitions and controlling changes to the system in a single location.

Distributed Configuration

Many item processing applications “push” the recognition engine to the branch server or workstation either via their own installation program or through a distributed installation program like SMS. These customers typically run branch capture and recognition at the same time prior to transmitting images to the central location. This approach utilizes distributed CPU’s that are available and can save the financial institution money in this area. However, implementing the proper controls to optimize recognition performance can be more challenging. Many times, item processing vendors have little reporting abilities to distinguish recognition rates from branch to branch. This makes managing performance more difficult. Additionally, when changing templates and OrboTool definitions, an automated program must be available to push the changes to each branch without manual intervention.

Lastly, high levels of automation and accuracy play an important part in this solution. The majority of financial institutions in this environment rely on centralized balancing and reject correction, so providing high levels of performance, i.e. 90-98% with accuracy levels of 99%+, streamlines balancing and minimizes the number of times checks need to be touched by a bank employee.