Image Processing and Binarisation for Camera OCR

Product-Line:
FlexiCapture Engine, FineReader Engine
Version:
10, 11
Type:
Technology & Features
Category:
Imaging

ABBYY OCR technologies contain Adaptive Binarisation
This technology is extremely important for high quality OCR results, because

  • Text should not be “lost” by applying wrong binarisation parameters
  • The characters should not “fall apart” or “glue together”

ABBYY core binarisation was re-designed improved in in V10 Technologies

  • Much less text is lost on low contrast images
  • Virtual lines(= “garbage”), because of shine though in the scanning process can be removed
    Important: Grey-scale images are required - it will not work with binary images only back-converted in grey-scale images ;-)

Image Preprocessing & Binarisation Technology have to be aligned, only then good OCR results can be archieved. ABBYY is optimizing, tuning and adjusting the core technologies in every major technology cycle.

Samples Camera OCR Image Pre-Processing & OCR result

3D Correction before applying OCR, delivers a more unified text result, because the differences in the fonts sizes are not “reconstructed”

Blured and unsharp images have to be sharpened before they are submitted to the OCR backend - the optimised ABBYY binarisation then will be able to read much more text from the images

When images are taken in low light conditions the camera sensors increase the sensitivity of the image sensor. This increases the so called ISO noise (= red, green, blue pixel-noise). This “digital dirt” then can result in a low quality binary image. So it is important to apply filters on the RGB image in a way that the binarisation can generate the best possible quality for text recognition.

Example

Proper image pre-processing has a big impact on the quality of the optical character recognition process (OCR) but also the source image quality affects the processing speed. The illustrations below was made with a pre-compiled code sample that comes with a FineReader Engine Trial.

On the left side the image was OCRed without any pre-processing, on the right side you can see the results with some pre-processing options turned on:

Results:

  • more characters were recognized (+150 %)
  • less of them were mark as uncertain (-73%)
  • processing is 3 times faster
This website uses cookies which enable you to see pages or use other functions of our websites. You can turn off such cookies in your browser’s settings. If you continue to use these pages, you consent to the use of cookies.