Signal models#

A concentration analysis (Concentration analysis) has two swappable mathematical pieces: a signal reduction that turns an image difference into a scalar, and a model that turns that scalar into a physical quantity.

Signal reduction#

MonochromaticReduction – take one colour channel ("red", "hue", …).
SignalReduction – general projection / dimension reduction.
AxisReduction, reduce_axis(), extrude_along_axis() – reduce or extend a spatial axis.

Models#

All models share the Model interface (__call__ maps signal to quantity) and compose with CombinedModel.

`LinearModel`, `ScalingModel`	linear / affine maps
`ClipModel`	clamp to a range
`StaticThresholdModel`,	fixed / data-driven thresholds
`DynamicThresholdModel`
`StandardOtsu`,	histogram-based thresholding
`TwoPeakHistogrammAnalysis` (and kin)
`KernelInterpolation`	non-linear map learned from labelled colour samples; handles curved colour paths
`ColorPathInterpolation`	interpolation along a colour path
`PWTransformation`	piecewise transformation
`HeterogeneousModel`,	apply a different model per region / facies
`HeterogeneousLinearModel`

BinaryDataSelector plus the criterion classes (ValueCriterion, GradientModulusCriterion, CombinedCriterion, …) build the masks that decide where a model is applied.

See also#