Illustration generated with AI; not experimental data.

Feature Extraction

JuSPICE includes an advanced feature extraction pipeline that computes per-pixel feature maps from image data for downstream analysis workflows. Every feature map is the same size as the input image (same height/width), so any combination of them can be stacked into one multi-channel array. Pick a feature below.

Gaussian

Multi-scale smoothing; the building block behind several other filters.

Sobel

Fast first-order gradient magnitude and direction.

Scharr

A rotationally accurate alternative to Sobel.

Laplace

Isotropic second-order edge/zero-crossing detector.

Hessian

Local curvature eigenvalues behind the ridge/vesselness filters.

Frangi

Vesselness filter for tube-like, elongated structures.

Meijering

Neuriteness filter for faint, thin curvilinear structures.

Sato

Tubeness filter that suppresses non-tubular responses.

Gabor

Orientation- and frequency-selective texture filters.

Local Binary Pattern (LBP)

Brightness-invariant local texture code.

Local Entropy

How random/textured a local neighborhood is.

Structure Tensor Coherence

How strongly and consistently oriented local structure is.

Core API

The primary entry point is spice.features.extract() (see FeatureExtractionAccessor), accessed directly on a SPICEData object. It returns a new SPICEData whose data array is a multi-channel feature stack, sharing the source object’s provenance lineage.

See juspice.feature_extract for full API details and Feature Extraction for a module-level explanation.

Notebook example

A complete workflow example is available in:

• notebooks/feature_extraction/feature_extract_basics.ipynb (see Notebook Examples)

The notebook uses load_data() and save_data() so each saved feature stack includes a metadata JSON sidecar and a reproducible history script.

References

Libraries

• scikit-image — provides all 16 filter implementations used in the feature pipeline (Gaussian, Sobel, Scharr, Laplace, Hessian, Frangi, Meijering, Sato, Gabor, LBP, Canny, entropy, structure tensor)

• NumPy — array stacking and statistical aggregation of feature maps

See Bibliography for the publications behind the methods on this page.