add_samples_image_class_gmmadd_samples_image_class_gmmAddSamplesImageClassGmmAddSamplesImageClassGmm (Operator)

Name

add_samples_image_class_gmmadd_samples_image_class_gmmAddSamplesImageClassGmmAddSamplesImageClassGmm — Add training samples from an image to the training data of a Gaussian Mixture Model.

Signature

add_samples_image_class_gmm(Image, ClassRegions : : GMMHandle, Randomize : )

add_samples_image_class_gmmadd_samples_image_class_gmmAddSamplesImageClassGmmAddSamplesImageClassGmmAddSamplesImageClassGmm adds training samples from the ImageImageImageImageimage to the Gaussian Mixture Model (GMM) given by GMMHandleGMMHandleGMMHandleGMMHandleGMMHandle. add_samples_image_class_gmmadd_samples_image_class_gmmAddSamplesImageClassGmmAddSamplesImageClassGmmAddSamplesImageClassGmm is used to store the training samples before a classifier to be used for the pixel classification of multichannel images with classify_image_class_gmmclassify_image_class_gmmClassifyImageClassGmmClassifyImageClassGmmClassifyImageClassGmm is trained. add_samples_image_class_gmmadd_samples_image_class_gmmAddSamplesImageClassGmmAddSamplesImageClassGmmAddSamplesImageClassGmm works analogously to add_sample_class_gmmadd_sample_class_gmmAddSampleClassGmmAddSampleClassGmmAddSampleClassGmm. The ImageImageImageImageimage must have a number of channels equal to NumDimNumDimNumDimNumDimnumDim, as specified with create_class_gmmcreate_class_gmmCreateClassGmmCreateClassGmmCreateClassGmm. The training regions for the NumClassesNumClassesNumClassesNumClassesnumClasses pixel classes are passed in ClassRegionsClassRegionsClassRegionsClassRegionsclassRegions. Hence, ClassRegionsClassRegionsClassRegionsClassRegionsclassRegions must be a tuple containing NumClassesNumClassesNumClassesNumClassesnumClasses regions. The order of the regions in ClassRegionsClassRegionsClassRegionsClassRegionsclassRegions determines the class of the pixels. If there are no samples for a particular class in ImageImageImageImageimage an empty region must be passed at the position of the class in ClassRegionsClassRegionsClassRegionsClassRegionsclassRegions. With this mechanism it is possible to use multiple images to add training samples for all relevant classes to the GMM by calling add_samples_image_class_gmmadd_samples_image_class_gmmAddSamplesImageClassGmmAddSamplesImageClassGmmAddSamplesImageClassGmm multiple times with the different images and suitably chosen regions. The regions in ClassRegionsClassRegionsClassRegionsClassRegionsclassRegions should contain representative training samples for the respective classes. Hence, they need not cover the entire image. The regions in ClassRegionsClassRegionsClassRegionsClassRegionsclassRegions should not overlap each other, because this would lead to the fact that in the training data the samples from the overlapping areas would be assigned to multiple classes, which may lead to a lower classification performance. Image data of integer type can be particularly badly suited for modeling with a GMM. RandomizeRandomizeRandomizeRandomizerandomize can be used to overcome this problem, as explained in add_sample_class_gmmadd_sample_class_gmmAddSampleClassGmmAddSampleClassGmmAddSampleClassGmm.

Execution Information

Multithreading type: reentrant (runs in parallel with non-exclusive operators).
Multithreading scope: global (may be called from any thread).
Processed without parallelization.

This operator modifies the state of the following input parameter:

GMMHandleGMMHandleGMMHandleGMMHandleGMMHandle

The value of this parameter may not be shared across multiple threads without external synchronization.

Parameters

ImageImageImageImageimage (input_object) (multichannel-)image → object (byte / cyclic / direction / int1 / int2 / uint2 / int4 / real)

Training image.

ClassRegionsClassRegionsClassRegionsClassRegionsclassRegions (input_object) region-array → object

Regions of the classes to be trained.

GMMHandleGMMHandleGMMHandleGMMHandleGMMHandle (input_control, state is modified) class_gmm → (integer)

GMM handle.

RandomizeRandomizeRandomizeRandomizerandomize (input_control) real → (real)

Standard deviation of the Gaussian noise added to the training data.

Default value: 0.0

Suggested values: 0.0, 1.5, 2.0

Restriction: Randomize >= 0.0

Result

If the parameters are valid, the operator add_samples_image_class_gmmadd_samples_image_class_gmmAddSamplesImageClassGmmAddSamplesImageClassGmmAddSamplesImageClassGmm returns the value 2 (H_MSG_TRUE). If necessary an exception is raised.