de.unihalle.informatik.MiToBo.enhance

Class LocallyAdaptiveContrastEnhancement

java.lang.Object

de.unihalle.informatik.Alida.operator.ALDOperator

de.unihalle.informatik.MiToBo.core.operator.MTBOperator

de.unihalle.informatik.MiToBo.enhance.LocallyAdaptiveContrastEnhancement

All Implemented Interfaces:

de.unihalle.informatik.Alida.datatypes.ALDConfigurationValidator, de.unihalle.informatik.Alida.operator.events.ALDOperatorExecutionProgressEventListener, EventListener

@ALDAOperator(genericExecutionMode=ALL,
              level=APPLICATION,
              shortDescription="Locally adaptive contrast enhancement for microscopy images. ")
public class LocallyAdaptiveContrastEnhancement
extends MTBOperator

This class implements contrast enhancement for microscopy images.

Reference:
Jyh-Ying Peng, Chun-Nan Hsu, Chung-Chih Lin,
"Adaptive Image Enhancement for Fluorescence Microscopy",
Int. Conf. on Technologies and Applications of Artificial Intelligence, pp. 9-16, 2010.

The basic idea of this algorithm is to enhance contrast by normalizing each pixel's intensity according to the standard intensity deviation in a local region around the pixel. Subsequently it should be easier to distinguish image background and relevant foreground structures from each other. Although in principal arbitrary segmentation methods could be applied after contrast enhancement, the algorithm is optimized for subsequent binarization, e.g. by Otsu thresholding. In addition, note that the algorithm does not work very well on images showing small structures on very noisy background (like P-bodies or stress granules). It is much better suited for larger structures like DAPI-stained nuclei which can more easily be distinguished from clutter, at least visually.

The size of each local region is adaptive and chosen so that the standard deviation in the region exceeds a certain fraction of the overall image standard deviation, where the fraction is usually lying in the range of 0.2 to 0.8. The ratio can be adapted by the corresponding parameter for the standard deviation ratio. The size of the local region is bounded to a maximum radius which is to be selected according to the size of the entities shown in the image.

The actual image enhancement is based on the following equations.
The radius for each pixel region is first of all calculated as follows:

\begin{equation*} r(x,y) = min \{ R_{max} , \underset{r}{min} \{ r > 0 | StD[R_r(x,y)] \geq T_{StD} \} \} \end{equation*}

where $R_{max}$ is the maximally allowed radius, $T_{StD}$ is the threshold for local standard deviation calculated as fraction of the image intensity standard deviation, and $StD[R_r(x,y)]$ is the standard deviation in a region $R_r(x,y)$ with radius $r$ around the current pixel $(x,y)$.

To avoid numerical instabilities the local standard deviation for normalization is calculated as follows, given the selected local region size $r^\ast$:

\begin{equation*} I_{StD}(x,y) = max \{ T_{StD} , StD[R_{r^\ast}(x,y)] \} \end{equation*}

Finally the normalization for contrast enhancement, i.e. the calculation of the new image intensity value $I_N(x,y)$ is done as follows:

\begin{eqnarray*} I_E(x,y) &=& E[R_{r^\ast}(x,y)] \\ I_{StD}(x,y) &=& StD[R_{r^\ast}(x,y)] \\ I_N(x,y) &=& \frac{I(x,y) - I_E(x,y)}{I_{StD}(x,y)} \end{eqnarray*}

As an extension to the original paper this operator features a mode for component-wise application of the algorithm. This means that the image is first of all thresholded and connected components are extracted. Then the contrast enhancement is applied to each component's bounding box separately. Finally, the result image is generated from all enhanced patches after they have been thresholded, i.e. the result image in this case is already a binary segmentation of foreground and background. This inherent binarization is done as the contrast-enhanced image does contain only fractions of reasonable information and, thus, is difficult to post-process without specific knowledge only available inside of this operator.

The basics of this operator have been implemented in the course of the bachelor thesis of Joachim Schumann in SS 2011.

Author:

Joachim Schumann, Birgit Moeller

Nested Class Summary

Nested classes/interfaces inherited from class de.unihalle.informatik.Alida.operator.ALDOperator

de.unihalle.informatik.Alida.operator.ALDOperator.HidingMode, de.unihalle.informatik.Alida.operator.ALDOperator.HistoryConstructionMode

Field Summary

Fields

Modifier and Type	Field and Description
private boolean	applyComponentWise Flag to activate component-wise application, yields a binary result image.
private boolean	calcRadiusImage Flag to enable calculation and display of radius image (only in non-component mode).
private MTBImage	inImg Input image to be processed.
private int	maxradius Maximal radius of region to be considered for intensity normalization.
private MTBImage	radiusImg Optional radius image.
private MTBImage	resultImg Enhanced or segmented result image, either of type double or binary.
private double	stdDevRatio Ratio of image standard deviation used as threshold for radius calculation.

Fields inherited from class de.unihalle.informatik.Alida.operator.ALDOperator

completeDAG, name, operatorExecutionEventlistenerList, portHashAccess, verbose, versionProvider

Constructor Summary

Constructors

Constructor and Description
LocallyAdaptiveContrastEnhancement() Default constructor.
LocallyAdaptiveContrastEnhancement(MTBImage inimg) Constructor with given image.
LocallyAdaptiveContrastEnhancement(MTBImage inimg, double pc, int maxr) Constructor with given image and parameters.

Method Summary

Modifier and Type	Method and Description
private static MTBImage[]	calcEnhancedImage(MTBImage img, double TstD, int maxradius, boolean calcRadiusImage) Calculates the new image intensity values of the image.
private static double	calcTStD(double StDI, double percentage) Calculates the threshold for the local standard daviation.
private static double	calStDI(MTBImage image) Calculates the standard deviation of the given image.
private MTBRegion2DSet	extractLocalComponents() Extracts list of connected components from Otsu thresholded input image.
String	getDocumentation()
MTBImage	getInputImage() Returns the input image, null if not set.
MTBImage	getRadiusImage() Returns supplemental radius image, null if not generated.
MTBImage	getResultImage() Returns result image.
boolean	isAppliedComponentwise() Return operator mode.
protected void	operate() This method does the actual work.
void	setInputImage(MTBImage inimg) Specify the input image.

Methods inherited from class de.unihalle.informatik.MiToBo.core.operator.MTBOperator

readResolve

Methods inherited from class de.unihalle.informatik.Alida.operator.ALDOperator

addOperatorExecutionProgressEventListener, addParameter, addParameter, addParameterUnconditioned, fieldContained, fireOperatorExecutionProgressEvent, getALDPortHashAccessKey, getConstructionMode, getHidingMode, getInactiveParameterNames, getInInoutNames, getInInoutNames, getInNames, getInOutNames, getMissingRequiredInputs, getName, getNumParameters, getOutInoutNames, getOutNames, getParameter, getParameterDescriptor, getParameterDescriptorUnconditioned, getParameterNames, getParameterUnconditioned, getSupplementalNames, getVerbose, getVersion, handleOperatorExecutionProgressEvent, hasInOutParameters, hasParameter, isAnnotatedParameter, isConfigured, print, print, print, printInterface, printInterface, readHistory, reinitializeParameterDescriptors, removeOperatorExecutionProgressEventListener, removeParameter, runOp, runOp, runOp, setConstructionMode, setConstructionMode, setConstructionMode, setHidingMode, setName, setParameter, setParameterUnconditioned, setVerbose, toStringVerbose, unconfiguredItems, validate, validateCustom, validateGeneric, writeHistory, writeHistory, writeHistory

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

inImg

@Parameter(label="Input Image",
           required=true,
           direction=IN,
           supplemental=false,
           description="Input image",
           mode=STANDARD)
private transient MTBImage inImg

Input image to be processed.

resultImg

@Parameter(label="Result Image",
           required=false,
           direction=OUT,
           supplemental=false,
           description="Result image")
private transient MTBImage resultImg

Enhanced or segmented result image, either of type double or binary.

Note that the enhanced, non-binarized image is scaled to [0,1].

maxradius

@Parameter(label="Maximal Region Radius",
           required=false,
           direction=IN,
           supplemental=false,
           dataIOOrder=-20,
           description="Max. Radius of Region",
           mode=ADVANCED)
private int maxradius

Maximal radius of region to be considered for intensity normalization.

This radius is supposed to be chosen according to the size of the entities of interest in the image.

stdDevRatio

@Parameter(label="Std. Deviation Ratio",
           required=false,
           direction=IN,
           mode=ADVANCED,
           dataIOOrder=-18,
           description="Percentage of StD used as threshold value.")
private double stdDevRatio

Ratio of image standard deviation used as threshold for radius calculation.

For each pixel the radius of the region used for intensity normalization is calculated as to be the radius of the smallest region showing a standard deviation that exceeds a threshold. This threshold is calculated as the given ratio of the image's standard deviation. The larger this ratio is set, the larger the regions will become in tendency.

applyComponentWise

@Parameter(label="Apply component-wise",
           required=false,
           direction=IN,
           mode=ADVANCED,
           description="Flag to enable component-wise processing, results in binary image.",
           dataIOOrder=-16)
private boolean applyComponentWise

Flag to activate component-wise application, yields a binary result image.

Usually for enhancing an image average intensity values and standard deviations are calculated over the complete image. But, if there are large regions without structure in an image, calculated values are probably misleading, resulting in image degradation rather than image enhancement.

One possibility to overcome this problem is to apply the intensity normalization only locally. In detail, the image is initially thresholded by Otsu, then dilated and connected components are extracted. Subsequently for each connected component the bounding box is extracted (slightly enlarged), and the normalization is only applied to the image region corresponding to the bounding box. At the end all enhanced regions are merged again to yield the final, enhanced image.

calcRadiusImage

@Parameter(label="Calculate Radius Image",
           required=false,
           direction=IN,
           mode=ADVANCED,
           supplemental=true,
           description="Flag for calculating radius image.")
private boolean calcRadiusImage

Flag to enable calculation and display of radius image (only in non-component mode).

radiusImg

@Parameter(label="Radius Image",
           required=false,
           direction=OUT,
           supplemental=true,
           mode=ADVANCED,
           description="Radius image")
private transient MTBImage radiusImg

Optional radius image.

Constructor Detail

LocallyAdaptiveContrastEnhancement

public LocallyAdaptiveContrastEnhancement()
                                   throws de.unihalle.informatik.Alida.exceptions.ALDOperatorException

Default constructor.

Throws:

de.unihalle.informatik.Alida.exceptions.ALDOperatorException

LocallyAdaptiveContrastEnhancement

public LocallyAdaptiveContrastEnhancement(MTBImage inimg)
                                   throws de.unihalle.informatik.Alida.exceptions.ALDOperatorException

Constructor with given image.

Parameters:

inimg - Input image.

Throws:

de.unihalle.informatik.Alida.exceptions.ALDOperatorException

LocallyAdaptiveContrastEnhancement

public LocallyAdaptiveContrastEnhancement(MTBImage inimg,
                                          double pc,
                                          int maxr)
                                   throws de.unihalle.informatik.Alida.exceptions.ALDOperatorException

Constructor with given image and parameters.

Parameters:

inimg - Input image.

Throws:

de.unihalle.informatik.Alida.exceptions.ALDOperatorException

Method Detail

setInputImage

public void setInputImage(MTBImage inimg)

Specify the input image.

Parameters:

inimg - Input image to be processed.

getInputImage

public MTBImage getInputImage()

Returns the input image, null if not set.

getResultImage

public MTBImage getResultImage()

Returns result image.

getRadiusImage

public MTBImage getRadiusImage()

Returns supplemental radius image, null if not generated.

isAppliedComponentwise

public boolean isAppliedComponentwise()

Return operator mode.

Returns:

If true, enhancement is applied component-wise.

operate

protected void operate()
                throws de.unihalle.informatik.Alida.exceptions.ALDOperatorException,
                       de.unihalle.informatik.Alida.exceptions.ALDProcessingDAGException

This method does the actual work.

Specified by:

operate in class de.unihalle.informatik.Alida.operator.ALDOperator

Throws:

de.unihalle.informatik.Alida.exceptions.ALDOperatorException

de.unihalle.informatik.Alida.exceptions.ALDProcessingDAGException

extractLocalComponents

private MTBRegion2DSet extractLocalComponents()
                                       throws de.unihalle.informatik.Alida.exceptions.ALDOperatorException,
                                              de.unihalle.informatik.Alida.exceptions.ALDProcessingDAGException

Extracts list of connected components from Otsu thresholded input image.

The method first applies Otsu thresholding to the given image and subsequently applies a dilation with a 3x3 structuring element. Then connected components are extracted, and finally all components with a size larger than 50 pixels are returned as result.

Returns:

List of connected components larger than 50 pixels.

Throws:

de.unihalle.informatik.Alida.exceptions.ALDOperatorException

de.unihalle.informatik.Alida.exceptions.ALDProcessingDAGException

calStDI

private static double calStDI(MTBImage image)

Calculates the standard deviation of the given image.

Parameters:

image - Image to process.

Returns:

Calculated standard deviation.

calcTStD

private static double calcTStD(double StDI,
                               double percentage)

Calculates the threshold for the local standard daviation.

Equation:

\begin{equation*} thresh = percentage \cdot StDI \end{equation*}

Parameters:

StDI - Standard deviation.

percentage - Specified percentage.

Returns:

Calculated threshold.

calcEnhancedImage

private static MTBImage[] calcEnhancedImage(MTBImage img,
                                            double TstD,
                                            int maxradius,
                                            boolean calcRadiusImage)

Calculates the new image intensity values of the image.

Parameters:

img - Image to process.

TstD - Standard deviation threshold to apply.

maxradius - Maximal local region radius.

calcRadiusImage - Flag to enable calculation of radius image.

Returns:

Array with enhanced image and (optionally) radius image.

getDocumentation

public String getDocumentation()

Overrides:

getDocumentation in class de.unihalle.informatik.Alida.operator.ALDOperator