Advanced Parameters

Segmentation Parameters

The following parameters are set and used in the process: Detecting the Characters.

Select the Segmentation Method

• UseSetSegmentationMethod ^{.NET Python} to select the algorithm used for segmentation:

Local segmentation (default):

• EOCR2SegmentationMethod_Local ^{.NET Python} is more complex and best suited for a non uniform background.

• Use SetMaxVariation ^{.NET Python} to define how sharp a blob in the image should be in order to be considered a potential character.

  • A region with clearly defined edges is generally considered stable while a blurry region is not.

  • Set this parameter between 0 and 1 (the default setting is 0.25).

  • A low setting allows only very stable blobs.

  • A high setting allows detection of blobs that are more unstable.

• Use SetDetectionDelta ^{.NET Python} to set the range of grayscale values used to determine the stability of a blob.

  • Set this parameter between 1 and 127 (the default setting is 12).

  • With a low setting, the algorithm is more sensitive to noise.

  • With a high setting, the algorithm is insensitive to blobs that have a low contrast with the background.

Global segmentation:

• EOCR2SegmentationMethod_Global ^{.NET Python} is faster and best suited for a clear background.

• It uses a simple threshold.

• Use SetGlobalSegmentationThresholdMode ^{.NET Python} to set the threshold computation method.

  • This parameter is a EThresholdMode ^{.NET Python}.

  • Use SetGlobalSegmentationAbsoluteThreshold ^{.NET Python} to set the threshold value when required.

• Use SetEnableSecondPassGlobalSegmentation ^{.NET Python} to perform the segmentation twice during the first pass.

  • It leads to more accurate results when the text background is not just plain.

  Warning

  This functionality is only available when GlobalSegmentationThresholdMode is set to the minimum residue (EThresholdMode_MinResidue ^{.NET Python}).

Local segmentation

Global segmentation with relative threshold set to 0.2.

Adjust the Fragmentation

• Use SetCharsMaxFragmentation ^{.NET Python} to set how small a blob can be for the segmentation algorithm to consider it as (part of) a character.

  • The minimum allowed area of a blob is given by:

  \[\text{minArea} = \text{CharsMaxFragmentation} × \text{CharsHeight} × min(\text{CharsWidthRange})\]

  • Set this parameter between 0 and 1 (the default setting is 0.1).

CharsMaxFragmentation = 0.1 (default):

CharsMaxFragmentation = 0.01:

Detection Parameters

The following parameters are set and used in the process: Detecting the Characters.

Select the Detection Method

If the topology is required, use SetDetectionMethod ^{.NET Python} to select the algorithm used for fitting:

For texts using a fixed width font (default):

• Use EOCR2DetectionMethod_FixedWidth ^{.NET Python}.

• Supports dotted text.

• Use SetRelativeSpacesWidthRange ^{.NET Python} to specify how wide the spaces between the words may be.

  • The box-fitting method tests the following range of spaces:

  \[ \begin{align}\begin{aligned}min(RelativeSpacesWidthRange) ×  min(charsWidthRange)\\≤ space ≤\\max(RelativeSpacesWidthRange) × max(charsWidthRange)\end{aligned}\end{align} \]

  • The upper bound of this parameter is also used when the topology is not required.

• Use SetCharsWidthBias ^{.NET Python} to add a bias towards wider or narrower character boxes.

• Use SetCharsSpacingBias ^{.NET Python} to add a bias towards a smaller or larger spacing between character boxes.

For texts using a proportional font:

• Use EOCR2DetectionMethod_Proportional ^{.NET Python}

• Use SetEnableCutLargeCharacter ^{.NET Python} so that the detection will try to cut wide segmented blobs into smaller boxes.

  • This parameter is also used when the topology is not required.

Warning

• With the fixed-width method, all the character boxes have the same width and they do not necessarily fit tightly around the characters.

• With the proportional method, the character boxes fit tightly around the characters. If any character falls outside the range of allowed widths, the detection fails.

Set the Text Rotation Angle

• Use SetTextAngleRange ^{.NET Python} to specify how the text in the image may be oriented. The box-fitting method tests the following range of rotation angles:

  \[ min(TextAngleRange) ≤ angle ≤ max(TextAngleRange)\]

  • The angles are defined with respect to the horizontal.

  • To set the unit for the angles (degrees, radians, revolutions or grades), use SetAngleUnit ^{.NET Python}.

  • The default setting is \([-20, 20]\) degrees.

  • This parameter is also used when the topology is not required.

Set a Second Detection Pass

• Use SetNumDetectionPasses ^{.NET Python} to configure a second pass for the textbox-fitting operation.

  • The first pass fits textboxes to all detected blobs.

  • The second pass selects only the blobs that are covered by the textboxes from the first pass. It fits textboxes to that subset of blobs, possibly resulting in a more optimal fit.

  • Set this parameter to either 1 or 2.

  • The default setting is 1.

  • This parameter is also used when the topology is not required.

NumDetectionPasses = 1:

The text angle estimate is slightly off.

NumDetectionPasses = 2:

The text angle estimate fits more closely.

No-Topology Mode Parameters

The following parameters are set and used in the process: Detect the Characters.

There are other parameters you can set if you are not using a topology:

Detect characters that are out of the size range:

• Use SetEnableOffSizeCharacter ^{.NET Python} to detect characters whose size (width and height) is out of the range but are in the vicinity of valid characters.

Group Blobs of a same character:

• Use SetRepasteObjects ^{.NET Python} to try and group blobs of the same character.

  • It can improve the detection of a clean segmentation with characters close to each other.

  • When true (default), the detection process tries to merge the blobs.