Default Feature Presets

Dragonfly's Machine Learning Segmentation module provides default feature presets for both pixel-based training and region-based training. You can use these presets as is, or edit them as required (see Editing Feature Presets). You can also create new presets (see Creating New Feature Presets).

Default Dataset Features Presets

The default dataset features presets listed below are available for creating new pixel-based or region-based classifiers (see Dataset Features for more information about dataset features).

You should note that the default dataset features presets are stored in the directory: \ProgramData\ORS\Dragonflyxx\Data\OrsTrainer\FeaturePresets.

It is usually better to edit the copy of a preset that is added to a model rather than the default preset itself.

Each filter in a dataset feature preset is applied separately during training and classification. The filters in a preset are not applied in a cascade or as a composite.

Dataset features presets
	Description	Used to…	Settings
Contrast		Enhance low-contrast elements.	Click to view the default settings
Extrema	Minimum and maximum value in a region.	Diminish the importance of small dark or light elements.	1. Maximum: Kernel = 2D circle of size 5. 2. Minimum: Kernel = 2D circle of size 5.
Gabors	Stack of Gabor filters with different angles.	Separate elements with oriented patterns.	Click to view the default settings
Local Binary Pattern	Local binary patterns computed with four different methods.	Separate elements with different textures. NOTE Not useful when applied with a histogram, since its output range is irrelevant.	Click to view the default settings
Moments	First, second, and third moments of the image.	Separate elements with different textures.	Click to view the default settings
Morphological	Morphological filters.	Uniformize intensity in a region, which can eliminate small areas within a region that can confuse a classifier.	Click to view the default settings
Neighbors	Stack of 3 x 3 editable filters with only one element of the kernel to 1. [1 0 0, 0 0 0, 0 0 0] [0 1 0, 0 0 0, 0 0 0] [0 0 1, 0 0 0, 0 0 0] [0 0 0, 1 0 0, 0 0 0] [0 0 0, 0 0 1, 0 0 0] etc...	Working on pixels only. To have information on surrounding pixels. Decrease the noise in the classification result.	Click to view the default settings
Self	The image itself.	Always. Might be redundant if you already have smoothing filters in your tree.	1. Arithmetic: A
Smoothing		Smooth noisy images, which helps a lot when working on pixels.	Click to view the default settings

Default Region Features Presets

The default region features presets listed below are available for creating new region-based classifiers (see Region Features for more information about region features).

You should note that the default region features presets are stored in the directory: \ProgramData\ORS\Dragonflyxx\Data\OrsTrainer\RegionFeaturePresets.

It is usually better to edit the copy of a preset that is added to a model rather than the default preset itself.

Region features presets
	Description	Useful when	Settings
Geometry	Geometrical characteristics of the regions.	You have regions big enough to embrace a complete object. The background is often very random in opposition with the elements to separate which have a more regular geometry.	Click to view the default settings
Histograms		Always. The number of bins is crucial, with an irrelevant number of bins, two classes could end up classified in the same bin.	1. Region Histogram: Bins: 32 2. Region Histogram: Bins: 16 3. Region Histogram: Bins: 8.
Neighbors Distance		You want to include or exclude dark regions in a bright surrounding or bright regions in a dark surrounding.	Click to view the default settings
Statistics		Always.	Click to view the default settings