TABLE OF CONTENTS
- HOW IT WORKS
- METHODS OF PROCESSING
- SUPPORTED IMAGE FORMATS
- RECOMMENDED IMAGE RESOLUTIONS
- SUPPORTED CHARACTER RESOLUTIONS
- SUPPORTED ANGLE RANGE OF IMAGES
- OPERATING SYSTEM-SPECIFIC DOCUMENTATIONS
- OCR ENGINE OUTPUT
- LICENSE ACTIVATION
- SUPPORT SERVICES
The ConPDS OCR engine is not an OCR SDK but a pre-compiled, plug-and-play library, ready-to-use in your software projects. It supports the detection of BIC container codes (ISO 6346), including ISO Size and Type and ILU container codes (EN 13044-1 standard).
HOW IT WORKS
- Capture the image with container code and load it into the OCR engine.
- Preprocessing is making image noise reduction, binarization, image emphasis, and skew correction.
- Detection is our optimized set of algorithms that analyze the image, detect areas with characters in patterns specified in BIC and ILU standards, cleanse the area, and remove any noise in each area.
- The recognition phase identifies and recognizes each character, combines them into a container code, and assigns a confidence factor the result. There can be multiple recognition results and a calculation of a confidence factor for each candidate done.
- Recognition results are returned as JSON responses and can be used in further processing or exported. The developer has full control over the OCR results.
METHODS OF PROCESSING
All versions of the ConPDS OCR Engine support two ways of processing images.
- Image file from storage
- Image binary from memory. It can be single frames from a video stream.
In both cases, it is up to the developer to capture images from a camera (mobile device, stationary camera, etc.) or capture frames from a video stream, e.g., by using RTSP.
SUPPORTED IMAGE FORMATS
Images for OCR processing should be in one of the following formats:
- JPEG format (https://en.wikipedia.org/wiki/JPEG).
- BMP format (https://en.wikipedia.org/wiki/BMP_file_format)
- PNG format (https://en.wikipedia.org/wiki/Portable_Network_Graphics)
RECOMMENDED IMAGE RESOLUTIONS
For the best and optimal recognition speed, we recommend below image resolutions.
Resolutions (HxW or WxH) *)
1024×768, 1280×960, 1400×1050, 1440×1080
1280×800, 1440×900, 1680×1050
*) Lower image resolutions are supported but will result in less accuracy in recognitions. Higher image resolutions are also supported, but as the image file size is big (2MB or higher), it will result in slow or even failed recognition.
The average OCR processing time for a JPEG image (1024x768 pixels, 120Kb in file size) is <100ms.
Invest some hours in optimizing the image capture process so that image resolutions are in recommended resolutions and aim for image file sizes not bigger than 200-250Kb. It will result in the fastest processing time and produce good recognition results.
SUPPORTED CHARACTER RESOLUTIONS
For optimal character recognition, we recommend the following character resolutions.
- 14 pixels – minimum *)
- 20 pixels – acceptable
- 50-100 pixels – recommended
- 200 pixels – maximum *)
*) Character resolution lower or higher will result in poor if any recognition.
SUPPORTED ANGLE RANGE OF IMAGES
The maximum image rotation – X (pitch), Y (yaw), Z (roll) - is ± 30° (recommended ± 15°).
An image rotation higher than will result in no recognition.