OCR Input Device
OCR - Scanning Resolution and Color Depth
Image scanning can be performed at different resolutions. Obviously, a higher resolution
allows the saving of more data but results in a larger image file. Although it is
generally accepted the 300 dpi scanning is the norm for human reading purposes, the
minimum requirement for OCR image scanning is 200 dpi. This is with the assumption that
the font size of the letters in the scanned document will have to be sufficiently large
to allow effective extraction. As such, if the document contains small print, a larger
resolution will be required.
The color depth of the scanned image also contributes to the accuracy of the recognition
process. Color depth affects file size very significantly. Fortunately, color depth is
not as relevant as scan resolution for OCR. The minimum requirement for OCR purposes
is bi-tonal scanning - i.e. black and white only.
For a good working system, scanning with 300 dpi and 256 grayscale is recommended.
Grayscale is recommended because although the words and letters in documents
for OCR are generally already of "black and white" origins, illumination plays
a big role
in creating a good scan. As such, the may be a need to perform adaptive
create a clean bi-tonal document image and adaptive thresholding
requires the scanning
to be performed in grayscale. A scanner that is able to create a
256-level grayscale scan
is sufficient mostly due to the fact that higher levels of
grayscale will be beyond the
capabilities of a human eye.
A good and high image resolution and color depth is almost always preferred to a lower
resolution and color for OCR purposes (of course without taking the file size
disadvantages into consideration). The difficult part is to test in order to determine
the best trade-off between definition and file size.
It is worthy to note that in most cases, the returns in going beyond 300 dpi and
256-level grayscale for OCR purposes do justify the file size trade-off. Similarly,
the use of color does not improve performance unless the text image to be scanned is in
a multi-color format. If the text is printed with a single color (not necessarily black),
and there is sufficient contrast with the paper background, grayscale scanners should
suffice to create a scan with sufficient character information. However, if multiple
colors are used extensively for letters and backgrounds, it may justify using a color
Note that an adaptive thresholding process will still need to be done in order to convert
to the image into a bi-tonal format for input into an appropriate OCR system.