Building Droids »

Enabling the Ecosystem is a series of interviews with those people and firms that are helping to build the overall Android ecosystem.

Today we wrap up the interview with Rick Pierson of ABBYY’s Mobility Group.

AG: How accurate are the recognitions?

RP: The accuracy rate depends on two factors; the camera and the photographer.  The camera has to have autofocus if you want to OCR magazine and business card sized print, the megapixels are not as important as the autofocus.  ABBYY does recommend a 2+ megapixel camera with autofocus.

The person taking the photograph does have to take a good quality picture, the better the photo the higher the OCR accuracy.

When I’m demoing one of ABBYY’s applications like business card reader on a Nokia n-95 (5 megapixels with autofocus) my recognition is usually 99% or better.

AG: How fast are those recognitions? After all, mobile phones do not have screaming-fast CPUs.

RP: This depends on the size of the image being OCR’ed and the quality of the picture.  On a Nokia N-95 it takes about 12 seconds once the photograph has been take to OCR the card, parse the information from the card and place it into the correct field in the handset contact list.  The OCR engine running on the handset was designed for smaller OCR applications; business card reader, poster, placard, paragraph from a magazine, etc.

ABBYY Mobile OCR Engine is a flexible and convenient technology that allows you to change its setting depending on the objective and the set of goals. You can select one of the two mobile recognition modes:

• Fast mode of express recognition – it is most convenient when the image is of good quality and allows cutting down the time required for its recognition and processing
• Full mode for accurate recognition – it is best for low-quality images, when more time is required to achieve the optimal result

The applications that will use the mobile OCR are not targeted for multiple full page OCR applications, if you want to build an application for that then we strongly suggest using ABBYY’s server based OCR solutions.

AG: If a developer wanted to deploy MOCR in their application, how much of a footprint will it require?

RP: The ABBYY Mobile OCR Engine is a compact code OCR technology and is optimized to work with small memory size systems such as mobile phones:

• ROM: 2.5 – 3Mb for program installation plus 0.5-1 Mb per recognition language; CJK requires 10 Mb per language.
• RAM: 2.5 – 3Mb for program storage plus 0.5-1 Mb for each recognition language used; 1.5 – 5Mb for program operation; 1 – 3Mb for storing the input picture/image file.

Exact memory requirements vary depending on the operating system and specific recognition tasks (e.g. multilingual recognition requires more memory).

We have a new algorithm for memory management which allows the system to determine the exact memory size required to process an image. This eliminates the necessity to allocate significant memory segments in advance, which has an impact on the recognition speed and the application’s capability to work reliably, ensuring efficiency and fast performance speed.

AG: Is it truly standalone, working on a disconnected device? Or does it communicate back to a server to perform the actual OCR function?

RP: The ABBYY Mobile OCR engine does the OCR on the handset without any connection to a network, it is truly stand alone.  ABBYY does however also offer a server based OCR solution, this is our “full” OCR engine called FineReader 9.  FineReader has more functionality however it is also 10 times the size of the Mobile version.