Similarity Measures

Similarity Measures. More...

Collaboration diagram for Similarity Measures:

Detailed Description

Similarity Measures.

All similarity measures derive from the SimilarityMeasureImpl class. The base class only provides the pointers to the two images that should be compared and some common properties like mask settings. The derived classes are required to implement the compute method that returns a actual similarity value. The ImageTransformer class is an abstraction of the transformation and provides methods to transform a pixel coordinate from the first image to the corresponding pixel coordinate of the second image. Therefore, to support new transformation models only the ImageTransformer has be updated, any class deriving from SimilarityMeasureImpl can remain unchanged. If no ImageTransformer is passed to compute, the images are compared directly without taking any transformations into account (also not the spacing). This should only be used with images of the same size. Any derived SimilarityMeasureImpl can be created directly or in a more generic way with the SimilarityMeasureFactory. The factory generates a SimilarityMeasureImpl from a Mode and a Device. While it is not enforced, it is encouraged to separate the implementations on different devices (e.g. CPU and GPU) into separate classes. This usually leads to a more robust implementation. Similarity measures should be re-created when the image size changes so that any intermediate buffers can be re-initialized.

There are several options to create a SimilarityMeasureImpl:

If only a specific implementation is required, it can be created directly. For example to create a Mutual Information similarity measure:

#include <ImFusion/Base/SimilarityMeasureMI.h>
...
  SimilarityMeasureMI* mi = new SimilarityMeasureImplMI_CPU(image1, image2);
  mi->compute(nullptr);
  delete mi;
...

In order to switch between different similarity measure, SimilarityMeasureFactory provides a generic interface based on properties. See SimilarityMeasureFactory for more information.

#include <ImFusion/Base/SimilarityMeasureFactory.h>
...
class SimExample
{
public:
  SimExample()
  {
  }
 
  void setParam(std::string name, int value)
  {
      // set a generic parameter of the SimilarityMeasure
      // in a Qt UI the ImFusion::PropertiesWidget can be used instead
      Properties p("");
      m_factory.similarityMeasureConfiguration(&p);
      p.setParam(name, value);
      m_factory.configureSimilarityMeasure(&p)
  }
 
  void compute()
  {
      SharedImage* img1 = ...;
      SharedImage* img2 = ...;
      // create SAD on CPU
      SimilarityMeasureImpl* sm = m_factory.generate(img1, img2);
      if(sm)
      {
          sm->compute(nullptr);
          delete sm;
      }
      // change the configuration to NCC with OpenGL
      m_factory.setMode(SimilarityMeasureFactory::NCC);
      m_factory.setDevice(Device_GL);
      sm = m_factory.generate(img1, img2);
      if(sm)
      {
          sm->compute(nullptr);
          delete sm;
      }
  }
 
private:
  SimilarityMeasureFactory m_factory;
};

When called the first time for a certain mode, similarityMeasureConfiguration will return the default parameters of the SimilarityMeasure.

If a persistent SimilarityMeasureImpl instance is required, the SimilarityMeasureWrapper can be used. However, this comes with less control when the actual instance is created and destroyed.

#include <ImFusion/Base/SimilarityMeasureWrapper.h>
...
class SimExample
{
public:
  void compute()
  {
      SharedImage* img1 = ...;
      SharedImage* img2 = ...;
      m_sm = new SimilarityMeasureWrapper(img1, img2, SimilarityMeasureFactory::SAD);
      m_sm->compute(nullptr); //implicitly creates SimilarityMeasureImpl instance
      m_sm->compute(nullptr); //uses the instance from before
      m_sm->setMode(SimilarityMeasureFactory::NCC); //implicitly deletes SimilarityMeasureImpl instance
      m_sm->setUseGPU(true);
      m_sm->compute(nullptr); //implicitly creates SimilarityMeasureImpl instance
      delete m_sm;
  }
private:
  SimilarityMeasureWrapper* m_sm;
};

Classes
class	SimilarityMeasureImpl
	Base class of a similarity measure between two images. More...
class	SimilarityMeasureDotProduct
class	SimilarityMeasureFactory
	The SimilarityMeasureFactory class provides a generic interface to any kind of SimilarityMeasure. More...
class	SimilarityMeasureImplHistBased
class	SimilarityMeasureLC2
class	SimilarityMeasureLNCC
class	SimilarityMeasureMI
	Computes the Mutual Information of two images. More...
class	SimilarityMeasureMI_CPU
	CPU implementation of Mutual Information. More...
class	SimilarityMeasureMI_GL
	GPU implementation of Mutual Information using modern OpenGL. More...
class	SimilarityMeasureML
	Machine Learning based similarity measure. More...
class	SimilarityMeasurePatchBased_GL
	Helper class for patch based similarity measures. More...
class	SimilarityMeasureImplPixelWise
class	SimilarityMeasurePixelWise_CPU
class	SimilarityMeasurePixelWise_GL
class	SimilarityMeasureWrapper
	Wrapper for safe and configurable use of various similarity measures. More...

Enumerations
enum	Device { Device_CPU , Device_GL }
	Enumeration for different computation devices.