| Customize Help

Guidelines for choosing context types


An OCR font context is a MIL object that stores the OCR font information, target image character size and spacing, constraints, and processing controls. When allocating your font context (using MocrAllocFont()), you can choose 1 of 2 types:

General OCR font context type

The M_GENERAL OCR font context type requires less information about the target string but works best on clean target images. If you need to specify the location of the string to eliminate erroneous results, use a child buffer to create a smaller area in which to search, using MbufChild...().

When using an M_GENERAL OCR font context, you should keep in mind the following conditions:

  • The target image should have a clearly visible threshold (binarization) between the characters and their background. The threshold must preserve the shape of the characters. Broken and/or touching characters can degrade the results.

  • A clean background of the target image. The cleaner the background, the better the results of a search.

  • The backgrounds should not have blobs of similar size and intensity as the characters.

If you are using clean target images but they are complex, have lighting variations, or require better binarization, you might consider trying the MIL String Reader module, which is more suitable for these cases. Note that under the correct conditions, OCR is typically faster.

The following MocrControl() control types are available only when using an M_GENERAL OCR font context:

An example image best suited to using an M_GENERAL OCR font context is:

If your image is greatly degraded or if you know a lot about the location of your target string(s), it is recommended to use the M_CONSTRAINED font context type.

Constrained OCR font context type

The M_CONSTRAINED OCR font context type works well with degraded target images. This type of context requires more information about the target string, but provides a more robust search. M_CONSTRAINED is best used when all the details are known about the target string. Especially, the more known about the target string's location and the size and spacing of its characters, the better the results of a search. Since more precise information is required to calibrate your font with that of the target string, you can automatically calibrate the font with the target string using MocrCalibrateFont(). Note that the automatic calibration of a font can only be done when using an M_CONSTRAINED OCR font context type.

An example image best suited to using an M_CONSTRAINED OCR font context is:

If your image contains a string comprised of unevenly spaced characters or a string that has blanks or broken characters, it is recommended to use the M_GENERAL OCR font context type.

Switching between the two

Defining a good OCR font context takes time. Modifying an existing OCR font context is faster than creating a new OCR font context from scratch. When dealing with two sets of images, for example, with similar OCR font requirements with regards to character representation, but differing in some other aspect (for example, sizing, spacing, processing controls, and/or quality), it might be faster to switch OCR font context types, reset the required constraints and controls, preprocess, and then read the new images. You can switch the type of context using the MocrControl() function with M_CONTEXT_CONVERT.

For example, after reading a series of images with an evenly spaced font (depicted below), to read a group of images where the font is not equally spaced, requires a switch from M_CONSTRAINED to M_GENERAL.

Note, when you switch between context types, any unsupported setting is reset to its default.

Deciding which OCR font context type to use

The following decision tree should help you decide when you should use an M_GENERAL or an M_CONSTRAINED OCR font context type. Note that, if a question does not apply to your situation, skip to the next question in the list.

  1. Can the target image be thresholded? That is, can a clear distinction be made between the background of the target image and the characters to be read?

  2. Does the target image have intensity problems; that is, does it have reflections or uneven lighting, or is it so heavily degraded that bits of the string are difficult to read with the naked eye?

  3. Is the target string made up of characters that vary in size? Does the target string contain variable spacing?

    Note that variable spacing does not include "jitter", which refers to tiny variations in the position of the target characters relative to the expected character position. Jitter can be corrected by calibrating your OCR font context (with MocrCalibrateFont()) to match the target string.

    Set TargetCharSizeXMin and TargetCharSizeYMin to measure to the clearest edge of the target characters to be read. Set TargetCharSizeXMax and TargetCharSizeYMax to measure to the faintest edge of the target character to be read - just before the shading (if any) matches the background color.

  4. Are any of the characters to be read broken?

    When there are broken characters, the best way to determine which OCR font context to use is to try both the M_GENERAL and the M_CONSTRAINED contexts. In a specific situation, one or the other can produce better results.

    If using an M_GENERAL OCR font context, use MocrControl() with M_BROKEN_CHAR set to M_ENABLE.

  5. Are any of the characters to be read touching?

    When there are touching characters, the best way to determine which OCR font context to use is to try both the M_GENERAL and the M_CONSTRAINED contexts. In a specific situation, one or the other can produce better results.

    If using an M_GENERAL OCR font context, use MocrControl() with M_TOUCHING_CHAR set to M_ENABLE.

If your answers were a combination of both M_GENERAL and M_CONSTRAINED OCR font contexts, use the OCR font context most often recommended. You will have to further manipulate your target image, or your OCR font (mostly using MocrControl()), to improve results.