CIB image toolbox technical manual

6. Image processing algorithms: Filters

6.2. Binarization filters

The following chapters explain each filter. Usually there is also an example given. The input image for the filters is the following picture:

COCOCLUST_Binarizer
FBCITB_Binarizer
FengBinarizerFilter
GaussianBlurFilter
LocalOtsuBinarizer
MedianBlurAGTFilter
MedianBlurATFilter
MedianBlurGATFilter
NativeAdaptiveBinarizer
NiblackBinarizerFilter
NICKBinarizerFilter
OtsuBinarizer
PureAdaptiveGaussianThresholdingFilter
PureAdaptiveThresholdingFilter
SauvolaBinarizerFilter
ThresholdFilter
WolfJolionBinarizerFilter

COCOCLUST_Binarizer

Parameters:

Threshold:45|CLAHEClipLimit:3.0|GaussianBlurKernelSize:19|CannyUpperThresholdCoeff:0.15

Output:

Implementation of COCOCLUST binarization.

This binarizator is described in the publication Kasar, T. and Ramakrishnan, A.G., 2009. COCOCLUST: Contour-based color clustering for robust binarization of colored text. Proc. The Third CBDAR, pp.11-17.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
<COCOCLUST_Binarizer> 
 <Threshold>45</Threshold>   
 <CLAHEClipLimit>3.0</CLAHEClipLimit>  
 <GaussianBlurKernelSize>19</GaussianBlurKernelSize> 
 <CannyUpperThresholdCoeff>0.15</CannyUpperThresholdCoeff> 
 <CannyLowerThresholdCoeff>0.05</CannyLowerThresholdCoeff> 
 <CannyMorphIters>4</CannyMorphIters>
</COCOCLUST_Binarizer>
...

</ProcessingProfile>

Property name	Description	Default value
Threshold	Threshold for preliminary clustering procedure	45
CLAHEClipLimit	Parameter for CLAHE procedure	3.0
GaussianBlurKernelSize	Parameter for Gaussian blur procedure	19
CannyUpperThresholdCoeff	Coefficient for upper threshold of Canny edge detector	0.15
CannyLowerThresholdCoeff	Coefficient for lower threshold of Canny edge detector	0.05
CannyMorphIters	Parameter of morphology operations	4

FBCITB_Binarizer

Parameters:

UseCanny:1|UseVariancesMap:1

Output:

Implementation of FBCITB binarization.

This binarizator is described in Kasar, T., Kumar, J. and Ramakrishnan, A.G., 2007, September. Font and background color independent text binarization. In Second international workshop on camera-based document analysis and recognition (pp. 3-9).

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
<FBCITB_Binarizer>  
 <UseCanny>1</UseCanny>  
 <UseVariancesMap>0</UseVariancesMap> 
 <VarianceMapThreshold>200</VarianceMapThreshold> 
 <CLAHEClipLimit>2.0</CLAHEClipLimit>  
 <GaussianBlurKernelSize>9</GaussianBlurKernelSize>   
 <CannyUpperThresholdCoeff>0.6</CannyUpperThresholdCoeff>  
 <CannyLowerThresholdCoeff>0.4</CannyLowerThresholdCoeff>  
 <BoundingRectangleMaxArea>0.3</BoundingRectangleMaxArea>
</FBCITB_Binarizer>
...
</ProcessingProfile>

Property name	Description	Default value
UseCanny	Is not equal 0 if Canny border detection is required	1
UseVariancesMap	Is not equal 0 variances map is used for contour detection	0
VarianceMapThreshold	Variance threshold	200
CLAHEClipLimit	Parameter for CLAHE procedure	2.0
GaussianBlurKernelSize	Parameter for Gaussian blur procedure	9
CannyUpperThresholdCoeff	Coefficient for upper threshold of Canny edge detector	0.6
CannyLowerThresholdCoeff	Coefficient for lower threshold of Canny edge detector	0.4
BoundingRectangleMaxArea	Coefficient for the bounding rectangle area	0.3

FengBinarizerFilter

Parameters:

Output:

Implementation of Feng binarization.

This binarizator is described in Khurshid, K., Siddiqi, I., Faure, C. and Vincent, N., 2009, January. Comparison of Niblack inspired binarization methods for ancient documents. In Document Recognition and Retrieval XVI (Vol. 7247, p. 72470U). International Society for Optics and Photonics.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
<FengBinarizerFilter>
 <WindowSize>21</WindowSize>
 <Alpha1>0.7</Alpha1>
 <k1>0.2</k1>
 <k2>0.03</k2>
 <Gamma>2.0</Gamma>
 <MorphIterationCount>2</MorphIterationCount>
</FengBinarizerFilter>
...
</ProcessingProfile>

Property name	Description	Default value
WindowSize	Size of sliding window	21
Alpha1	Coefficient for threshold calculation	0.7
k1	Coefficient for threshold calculation	0.2
k2	Coefficient for threshold calculation	0.03
Gamma	Coefficient for threshold calculation	2.0
MorphIterationCount	Count of iterations of final dilate and erode operations	2

GaussianBlurFilter

In image processing, a Gaussian blur is the result of blurring an image by a Gaussian function. This function used to reduce Image noise and detail.

Parameters of this filter can be described in profile:

Please note that “KernelSize” only accept odd numbers!

<?xml version="1.0" encoding="utf-8"?>
<algorithms>
<algorithm name="ImageManipulation" TracingEnabled="false" inputs="in1=$1" outputs="out4=$1">
<layer>
<processor params="KernelSize:21" type="GaussianBlurFilter" outputs="out4" inputs="in1" />
</layer>
</algorithm>
</algorithms>

LocalOtsuBinarizer

Parameters:

Output:

Implementation of Otsu binarization algorithm.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 <LocalOtsuBinarizer>
  <MaxValue>255.0</MaxValue>
  <CLAHEClipLimit>0.0</CLAHEClipLimit> 
  <GaussianBlurKernelSize>19</GaussianBlurKernelSize>  
  <CannyUpperThresholdCoeff>0.15</CannyUpperThresholdCoeff>
  <CannyLowerThresholdCoeff>0.01</CannyLowerThresholdCoeff>  
  <CannyMorphIters>1</CannyMorphIters>
  </LocalOtsuBinarizer>
...
</ProcessingProfile>

Property name	Description	Default value
MaxValue	Maximal value	255.0
CLAHEClipLimit	Parameter for CLAHE procedure	0.0
GaussianBlurKernelSize	Parameter for Gaussian blur procedure	19
CannyUpperThresholdCoeff	Coefficient for upper threshold of Canny edge detector	0.15
CannyLowerThresholdCoeff	Coefficient for lower threshold of Canny edge detector	0.01
CannyMorphIters	Parameter of morphology operations	1

MedianBlurAGTFilter

Parameters:

KernelSize:3|BlockSize:5|Shift:0.0|MaxValue:255.0

Output:

Median blur + Adaptive Gaussian threshold filter.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 <MedianBlurAGTFilter>
  <KernelSize>3</KernelSize>  
  <MaxValue>255.0</MaxValue> 
  <BlockSize>5</BlockSize>
  <Shift>0.0</Shift>
 </MedianBlurAGTFilter>
...
</ProcessingProfile>

Property name	Description	Default value
KernelSize	Size of kernel of median blur filter	3
BlockSize	Size of sliding window	5
Shift	Constant subtracted from the mean or weighted mean	0.0
MaxValue	Color of background of binarized image	255.0

MedianBlurATFilter

Parameters:

KernelSize:3|BlockSize:5|Shift:0.0|MaxValue:255.0

Output:

Median blur + Adaptive threshold filter.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 <MedianBlurATFilter>  
  <KernelSize>3</KernelSize>
  <MaxValue>255.0</MaxValue> 
  <BlockSize>5</BlockSize> 
  <Shift>0.0</Shift>
 </MedianBlurATFilter>
...
</ProcessingProfile>

Property name	Description	Default value
KernelSize	Size of kernel of median blur filter	3
BlockSize	Size of sliding window	5
Shift	Constant subtracted from the mean or weighted mean	0.0
MaxValue	Color of background of binarized image	255.0

MedianBlurGATFilter

Parameters:

Default Parameters

Output:

Gaussian blur + Adaptive threshold filter.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 <MedianBlurGATFilter>
  <KernelSize>3</KernelSize>
  <SigmaX>10.0</SigmaX>
  <SigmaY>0.0</SigmaY>
  <MaxValue>255.0</MaxValue>
  <BlockSize>5</BlockSize>
  <Shift>0.0</Shift>
 </MedianBlurGATFilter>
...
</ProcessingProfile>

Property name	Description	Default value
KernelSize	Size of kernel of median blur filter	3
SigmaX	Sigma for horizontal direction	10.0
SigmaY	Sigma for vertical direction	0.0
MaxValue	Color of background of binarized image	255.0
BlockSize	Size of sliding window	5
Shift	Constant subtracted from the mean or weighted mean	0.0

NativeAdaptiveBinarizer

Parameters:

Default Parameters

Output:

Native adaptive thresholding.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 <NativeAdaptiveBinarizer>  
  <IsGaussianBlurReqiured>0</IsGaussianBlurReqiured> 
  <MedianBlurKernelSize>5</MedianBlurKernelSize>
  <GaussianBlurKernelSize>3</GaussianBlurKernelSize>
  <GaussianBlurSigma>150</GaussianBlurSigma
  <IsAdaptiveThresholdCalculatedByGaussian>1</IsAdaptiveThresholdCalculatedByGaussian>
  <AdaptiveThresholdingMaxValue>255.0</AdaptiveThresholdingMaxValue>
  <AdaptiveThresholdingBlockSize>19</AdaptiveThresholdingBlockSize>
  <AdaptiveThresholdingShift>8</AdaptiveThresholdingShift>
  <BilateralFilterBlockSize>5</BilateralFilterBlockSize>
  <BilateralFilterColorSigma>50.0</BilateralFilterColorSigma>
  <BilateralFilterSpaceSigma>50.0</BilateralFilterSpaceSigma>
 </NativeAdaptiveBinarizer>
...
</ProcessingProfile>

Property name	Description	Default value
IsGaussianBlurReqiured	Flag of usage Gaussian blur	0
MedianBlurKernelSize	Kernel size for median blur	5
GaussianBlurKernelSize	Kernel size for Gaussian blur	3
GaussianBlurSigma	Gaussian blur sigma	150
IsAdaptiveThresholdCalculatedByGaussian	Flag of base of adaptive thresholding type	1
AdaptiveThresholdingMaxValue	New value for pixel which intensity greater than threshold value	255.0
AdaptiveThresholdingBlockSize	Kernel size for adaptive thresholding procedure	19
AdaptiveThresholdingShift	Shifting value for adaptive thresholding procedure	8
BilateralFilterBlockSize	Kernel size for bilateral filtration. If less than 0 then filtration isn't used	5
BilateralFilterColorSigma	Sigma for intensity range	50.0
BilateralFilterSpaceSigma	Sigma for space range	50.0

NiblackBinarizerFilter

Parameters:

WindowSize:101|ThresholdCoefficient:0.01|MorphIterationCount:2

Output:

Implementation of Niblack binarization.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 
<NiblackBinarizerFilter>  
 <WindowSize>101</WindowSize> 
 <ThresholdCoefficient>0.01</ThresholdCoefficient> 
 <MorphIterationCount>2</MorphIterationCount>
 </NiblackBinarizerFilter>
...
</ProcessingProfile>

Property name	Description	Default value
WindowSize	Size of sliding window	101
ThresholdCoefficient	Coefficient $k$ for threshold calculation	0.01
MorphIterationCount	Count of iterations of final dilate and erode operations	2

NICKBinarizerFilter

Parameters:

WindowSize:101|ThresholdCoefficient:0.01|MorphIterationCount:2

Output:

Implementation of NICK binarization.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
<NICKBinarizerFilter>  
 <WindowSize>101</WindowSize>  
 <ThresholdCoefficient>0.01</ThresholdCoefficient>
 <MorphIterationCount>2</MorphIterationCount>
 </NICKBinarizerFilter>
...
</ProcessingProfile>

Property name	Description	Default value
WindowSize	Size of sliding window	101
ThresholdCoefficient	Coefficient $k$ for threshold calculation	0.01
MorphIterationCount	Count of iterations of final dilate and erode operations	2

OtsuBinarizer

Parameters:

MaxValue:255

Output:

Interface for Otsu binarization algorithm.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 <OtsuBinarizer> 
  <MaxValue>255.0</MaxValue>
  </OtsuBinarizer>
...
</ProcessingProfile>

Property name	Description	Default value
MaxValue	Color of background of binarized image	255.0

PureAdaptiveGaussianThresholdingFilter

Parameters:

MaxValue:255|BlockSize:5|Shift:0.0

Output:

Pure adaptive Gaussian thresholding filter.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 <PureAdaptiveGaussianThresholdingFilter> 
  <MaxValue>255.0</MaxValue>
  <BlockSize>5</BlockSize> 
  <Shift>0.0</Shift>
 </PureAdaptiveGaussianThresholdingFilter>
...
</ProcessingProfile>

Property name	Description	Default value
MaxValue	Color of background of binarized image	255.0
BlockSize	Size of sliding window	5
Shift	Constant subtracted from the mean or weighted mean	0.0

PureAdaptiveThresholdingFilter

Parameters:

MaxValue:255|BlockSize:5|Shift:0.0

Output:

Pure adaptive thresholding filter.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 <PureAdaptiveThresholdingFilter> 
  <MaxValue>255.0</MaxValue>
  <BlockSize>5</BlockSize>
  <Shift>0.0</Shift>
 </PureAdaptiveThresholdingFilter>
...
</ProcessingProfile>

Property name	Description	Default value
MaxValue	Color of background of binarized image	255.0
BlockSize	Size of sliding window	5
Shift	Constant subtracted from the mean or weighted mean	0.0

SauvolaBinarizerFilter

Parameters:

WindowSize:101|ThresholdCoefficient:0.00|MorphIterationCount:2

Output:

Implementation of Sauvola binarization.

This binarizator is described in Shafait, F., Keysers, D. and Breuel, T.M., 2008, January. Efficient implementation of local adaptive thresholding techniques using integral images. In Document recognition and retrieval XV (Vol. 6815, p. 681510). International Society for Optics and Photonics.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
<SauvolaBinarizerFilter> 
 <WindowSize>101</WindowSize>  
 <ThresholdCoefficient>0.01</ThresholdCoefficient>
 <MorphIterationCount>2</MorphIterationCount>
</SauvolaBinarizerFilter>
...
</ProcessingProfile>

Property name	Description	Default value
WindowSize	Size of sliding window	101
ThresholdCoefficient	Coefficient for threshold calculation	0.01
MorphIterationCount	Count of iterations of final dilate and erode operations	2

ThresholdFilter

Parameters:

ThresholdValues:100;128;90

Output:

Pure thresholding binarizator.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 <ThresholdFilter>
   
<ThresholdValues>100;128;90</ThresholdValues>
 </ThresholdFilter>
...
</ProcessingProfile>

Property name	Description	Default value
ThresholdValues	Values of threshold for each color channel	100; 128; 90

Also we can define equal value of thresholds for each channel:

...

</ThresholdFilter>

...

</ProcessingProfile>

WolfJolionBinarizerFilter

Parameters:

WindowSize:101|ThresholdCoefficient:0.00|MorphIterationCount:2

Output:

Implementation of WolfJolion binarization.

Parameters of this filter can be described in profile:

<ProcessingProfile>
...
 
<WolfJolionBinarizerFilter> 
 <WindowSize>101</WindowSize> 
 <ThresholdCoefficient>0.01</ThresholdCoefficient>
 <MorphIterationCount>2</MorphIterationCount>
 </WolfJolionBinarizerFilter>
...
</ProcessingProfile>

Property name	Description	Default value
WindowSize	Size of sliding window	101
ThresholdCoefficient	Coefficient for threshold calculation	0.01
MorphIterationCount	Count of iterations of final dilate and erode operations	2