ilastik Documentation Object Classification Features

Object Classification Features

ilastik object features describe objects in terms of numbers. These are used in classification to differentiate between different types of objects (classes). Per default ilastik comes with 3 feature plugins: “Standard Object Features”, “Skeleton Features” (2D only), and “Convex Hull Features”.

A short overview of available object features is given in the following segment of the webinar ilastik beyond pixel classification - [NEUBIASAcademy@Home]:

Some practical advice on selecting features can be found in our i2k ilastik tutorial:

Following here is a list of all available object features along with their description.

Standard Object Features

Bounding Box Maximum
The coordinates of the upper right corner of the object's bounding box. The first axis is x, then y, then z (if available).
Bounding Box Minimum
The coordinates of the lower left corner of the object's bounding box. The first axis is x, then y, then z (if available).
Size in pixels
Total size of the object in pixels. No correction for anisotropic resolution or anything else.
Covariance of Channel Intensity
For multi-channel images this feature computes the covariance between the channels inside the object.
Covariance of Channel Intensity in neighborhood
For multi-channel images this feature computes the covariance between the channels in the object neighborhood. The size of the neighborhood is determined from the controls in the lower part of the dialogue.
Histogram of Intensity
Histogram of the intensity distribution inside the object. The histogram has 64 bins and its range is computed from the global minimum and maximum intensity values in the whole image.
Histogram of Intensity in neighborhood
Histogram of the intensity distribution in the object neighborhood. The histogram has 64 bins and its range is computed from the global minimum and maximum intensity values in the whole image. The size of the neighborhood is determined from the controls in the lower part of the dialogue.
Kurtosis of Intensity
Kurtosis of the intensity distribution inside the object, also known as the fourth standardized moment. This feature measures the heaviness of the tails for the distribution of intensity over the object's pixels. For multi-channel data, this feature is computed channel-wise. If all pixels in an object have the same value, you may encounter a 'bad features' warning when computing Kurtosis. Kurtosis will have a value of 0 for these objects.
Kurtosis of Intensity in neighborhood
Kurtosis of the intensity distribution in the object neighborhood, also known as the fourth standardized moment. This feature measures the heaviness of the tails for the distribution of intensity over the object's pixels. For multi-channel data, this feature is computed channel-wise. If all pixels in an object have the same value, you may encounter a 'bad features' warning when computing Kurtosis. Kurtosis will have a value of 0 for these objects. The size of the neighborhood is determined from the controls in the lower part of the dialogue.
Maximum intensity
Maximum intensity value inside the object. For multi-channel data, this feature is computed channel-wise.
Maximum intensity in neighborhood
Maximum intensity value in the object neighborhood. For multi-channel data, this feature is computed channel-wise. The size of the neighborhood is determined from the controls in the lower part of the dialogue.
Mean Intensity
Mean intensity inside the object. For multi-channel data, this feature is computed channel-wise.
Mean Intensity in neighborhood
Mean intensity in the object neighborhood. For multi-channel data, this feature is computed channel-wise. The size of the neighborhood is determined from the controls in the lower part of the dialogue.
Minimum intensity
Minimum intensity value inside the object. For multi-channel data, this feature is computed channel-wise.
Minimum intensity in neighborhood
Minimum intensity value in the object neighborhood. For multi-channel data, this feature is computed channel-wise. The size of the neighborhood is determined from the controls in the lower part of the dialogue.
PrincipalAxes
PrincipalAxes, stay tuned for more details
Quantiles of Intensity
Quantiles of the intensity distribution inside the object, in the following order: 0%, 10%, 25%, 50%, 75%, 90%, 100%.
Principal components of the object
Eigenvectors of the PCA on the coordinates of the object's pixels. Very roughly, this corresponds to the axes of an ellipse fit to the object. The axes are ordered starting from the one with the largest eigenvalue.
Center of the object
Average of the coordinates of this object's pixels.
Radii of the object
Eigenvalues of the PCA on the coordinates of the object's pixels. Very roughly, this corresponds to the radii of an ellipse fit to the object. The radii are ordered, with the largest value as first.
Skewness of Intensity
Skewness of the intensity distribution inside the object, also known as the third standardized moment. This feature measures the asymmetry of the intensity distribution inside the object. For multi-channel data, this feature is computed channel-wise. If all pixels in an object have the same value, you may encounter a 'bad features' warning when computing Skewness. Skewness will have a value of 0 for these objects.
Skewness of Intensity in neighborhood
Skewness of the intensity distribution in the object neighborhood, also known as the third standardized moment. This feature measures the asymmetry of the intensity distribution in the object neighborhood. For multi-channel data, this feature is computed channel-wise. If all pixels in an object have the same value, you may encounter a 'bad features' warning when computing Skewness. Skewness will have a value of 0 for these objects. The size of the neighborhood is determined from the controls in the lower part of the dialogue.
Total Intensity
Sum of intensity values for all the pixels inside the object. For multi-channel images, computed channel-wise.
Total Intensity in neighborhood
Sum of intensity values for all the pixels in the object neighborhood. For multi-channel images, computed channel-wise. The size of the neighborhood is determined from the controls in the lower part of the dialogue.
Variance of Intensity
Variance of the intensity distribution inside the object. For multi-channel data, this feature is computed channel-wise.
Variance of Intensity in neighborhood
Variance of the intensity distribution in the object neighborhood. For multi-channel data, this feature is computed channel-wise. The size of the neighborhood is determined from the controls in the lower part of the dialogue.

Convex Hull Features

Convexity
The ratio between the areas of the object and its convex hull (<= 1)
Defect Center
Combined centroid of convexity defects, which are defined as areas of the convex hull, not covered by the original object.
Number of Defects
Total number of defects, i.e. number of connected components in the area of the convex hull, not covered by the original object
Mean Defect Displacement
Mean distance between the centroids of the original object and the centroids of the defects, weighted by defect area.
Kurtosis of Defect Area
Kurtosis (4th standardized moment, measure of tails' heaviness) of the distribution of the areas of convexity defects. Defects are defined as connected components in the area of the convex hull, not covered by the original object.
Mean Defect Area
Average of the areas of convexity defects. Defects are defined as connected components in the area of the convex hull, not covered by the original object.
Skewness of Defect Area
Skewness (3rd standardized moment, measure of asymmetry) of the distribution of the areas of convexity defects. Defects are defined as connected components in the area of the convex hull, not covered by the original object.
Variance of Defect Area
Variance of the distribution of areas of convexity defects. Defects are defined as connected components in the area of the convex hull, not covered by the original object.
Convex Hull Center
Centroid of the convex hull of this object. The axes order is x, y, z
Convex Hull Area
Area of the convex hull of this object
Object Center
Centroid of this object. The axes order is x, y, z
Object Area
Area of this object, computed from the interpixel contour (can be slightly larger than simple size of the object in pixels). This feature is used to compute convexity.

Skeleton Features

Average Branch Length
Average length of a branch in the skeleton
Number of Branches
Total number of branches in the skeleton of this object.
Diameter
The longest path between two endpoints on the skeleton.
Euclidean Diameter
The Euclidean distance between the endpoints (terminals) of the longest path on the skeleton
Number of Holes
The number of cycles in the skeleton (i.e. the number of cavities in the region)
Center of the Skeleton
The coordinates of the midpoint on the longest path between the endpoints of the skeleton.
Length of the Skeleton
Total length of the skeleton in pixels